Low-dose metformin targets the lysosomal AMPK pathway through PEN2

Metformin, the most prescribed antidiabetic medicine, has shown other benefits such as anti-ageing and anticancer effects1–4. For clinical doses of metformin, AMP-activated protein kinase (AMPK) has a major role in its mechanism of action4,5; however, the direct molecular target of metformin remains unknown. Here we show that clinically relevant concentrations of metformin inhibit the lysosomal proton pump v-ATPase, which is a central node for AMPK activation following glucose starvation6. We synthesize a photoactive metformin probe and identify PEN2, a subunit of γ-secretase7, as a binding partner of metformin with a dissociation constant at micromolar levels. Metformin-bound PEN2 forms a complex with ATP6AP1, a subunit of the v-ATPase8, which leads to the inhibition of v-ATPase and the activation of AMPK without effects on cellular AMP levels. Knockout of PEN2 or re-introduction of a PEN2 mutant that does not bind ATP6AP1 blunts AMPK activation. In vivo, liver-specific knockout of Pen2 abolishes metformin-mediated reduction of hepatic fat content, whereas intestine-specific knockout of Pen2 impairs its glucose-lowering effects. Furthermore, knockdown of pen-2 in Caenorhabditis elegans abrogates metformin-induced extension of lifespan. Together, these findings reveal that metformin binds PEN2 and initiates a signalling route that intersects, through ATP6AP1, the lysosomal glucose-sensing pathway for AMPK activation. This ensures that metformin exerts its therapeutic benefits in patients without substantial adverse effects. The molecular target of the antidiabetic medicine metformin is identified as PEN2, a subunit of γ-secretases, and the PEN2–ATP6AP1 axis offers potential targets for screening for metformin substitutes.

Metformin, the most prescribed antidiabetic medicine, has shown other benefits such as anti-ageing and anticancer effects [1][2][3][4] . For clinical doses of metformin, AMP-activated protein kinase (AMPK) has a major role in its mechanism of action 4,5 ; however, the direct molecular target of metformin remains unknown. Here we show that clinically relevant concentrations of metformin inhibit the lysosomal proton pump v-ATPase, which is a central node for AMPK activation following glucose starvation 6 . We synthesize a photoactive metformin probe and identify PEN2, a subunit of γ-secretase 7 , as a binding partner of metformin with a dissociation constant at micromolar levels. Metformin-bound PEN2 forms a complex with ATP6AP1, a subunit of the v-ATPase 8 , which leads to the inhibition of v-ATPase and the activation of AMPK without effects on cellular AMP levels. Knockout of PEN2 or re-introduction of a PEN2 mutant that does not bind ATP6AP1 blunts AMPK activation. In vivo, liver-specific knockout of Pen2 abolishes metformin-mediated reduction of hepatic fat content, whereas intestine-specific knockout of Pen2 impairs its glucose-lowering effects. Furthermore, knockdown of pen-2 in Caenorhabditis elegans abrogates metformininduced extension of lifespan. Together, these findings reveal that metformin binds PEN2 and initiates a signalling route that intersects, through ATP6AP1, the lysosomal glucose-sensing pathway for AMPK activation. This ensures that metformin exerts its therapeutic benefits in patients without substantial adverse effects.
Metformin is the usual first-line drug of choice to reduce blood glucose levels in patients with type 2 diabetes mellitus. It also has other clinically beneficial effects such as reductions in body weight and hepatic fat content, and decreased cancer incidence in patients with diabetes who take the drug 1,3 . Administration of metformin to various organisms, including nematodes (C. elegans) and mice, can also extend lifespan and health span 9,10 . Metformin requires transporters of the OCT family to enter cells, which restricts its primary target organs to the liver, the kidney and the intestine 5,11 . Various mechanisms of action for metformin to exert its roles have been proposed. Metformin can inhibit complex I of the mitochondrial electron transport chain in hepatocytes 12,13 , which leads to decreases in ATP and increases in AMP levels and in turn activates AMPK through the canonical adenine-nucleotide-dependent mechanism 14 . Increased AMP also inhibits fructose-1,6-bisphosphatase-1 and adenylate cyclase to block gluconeogenesis 15,16 . Metformin has also been proposed to alter cellular redox status, which increases NAD + /NADH ratios and leads to the suppression of the utilization of gluconeogenic substrates. Metformin may also exert its glucose-lowering effects in the gut by promoting the secretion of glucagon-like peptide 1 (GLP-1) 1 .
Among the various potential effectors of metformin identified, AMPK, a master controller of metabolic homeostasis, has been placed at centre stage 17,18 . AMPK, through phosphorylating acetyl-CoA carboxylase 1 (ACC1) and ACC2, is indispensable for the attenuation of hepatic steatosis and atherosclerosis in diabetic mice that have been given chronic metformin treatment 19,20 . Duodenal activation of AMPK is essential for GLP-1 secretion in L cells, and is required for the acute glucose-lowering effect of metformin when orally administered 21 . Furthermore, the metformin-mediated retardation of ageing in C. elegans is through an AMPK-dependent mechanism 9,18 .
It has been widely accepted that metformin activates AMPK by inhibiting complex I of the mitochondrial electron transport chain, which impairs ATP synthesis and in turn increases AMP/ATP and ADP/ ATP ratios [12][13][14] . However, the decrease in energy levels could only be observed at peak concentrations after high doses of metformin in mice (≥ 250 mg kg -1 orally, which yields peak plasma concentrations of Article 125-150 μM after 1-2 h and rapidly decreases thereafter 16 ). By comparison, the plasma metformin concentrations in patients taking standard clinical doses of 1.5-2 g per day (Glucophage, 0.5 g three times a day or four times a day) have been reported to be only 5-30 μM (ref. 11 ) (Extended Data Fig. 1a), which may not be sufficient to increase AMP/ ATP and ADP/ATP ratios 22,23 . Therefore, it is necessary to explore how clinically relevant doses of metformin activates AMPK.

PEN2 binds to metformin
We found that metformin at clinical doses sufficiently inhibited the vacuolar H + -ATPase (v-ATPase) on the lysosome (Fig. 1a, b and Extended Data Fig. 1, with detailed discussions in Supplementary Note 1). We therefore used an affinity-based approach to analyse protein extracts of purified lysosomes to identify potential direct targets for metformin (Fig. 2a). Two types of photoactive metformin probes, Met-P1 and Met-P2, were synthesized (Extended Data Fig. 2a), but only Met-P1 was able to inhibit lysosomal acidification; Met-P2 had no effect and was therefore discarded (Extended Data Fig. 2b). After incubation with lysosome lysates, Met-P1 was conjugated to proteins by ultraviolet irradiation and then biotinylated (chemical reactions shown in Extended Data Fig. 2c). NeutrAvidin beads were used to pull down the conjugates for analysis by mass spectrometry (MS). As listed in Supplementary  Table 1, we engineered expression plasmids for a total of 367 proteins, and verified that 113 proteins of them could be pulled down by Met-P1 when individually expressed in HEK293T cells (Supplementary Note 2). Next, we individually knocked down those 113 proteins in mouse embryonic fibroblasts (MEFs) through lentivirus-mediated short hairpin RNA (shRNA) silencing. We observed that depletion of PEN2, but not others, rendered the cells insensitive to metformin treatment, as assessed by levels of AMPK activation and inhibition of v-ATPase (Extended Data Figs. 2d and 3a, b). Consistently, knockout of PEN2 blocked low-dose metformin-induced AMPK activation and v-ATPase inhibition in primary hepatocytes, MEFs and HEK293T cells (Fig. 2b, c and Extended Data Fig. 3c-i, k, l; note that knockout of PEN2 did not affect basal lysosomal pH levels (Supplementary Note 2)). Of note, depletion of PEN2 in all three cell types did not affect the transport of metformin into cells (Extended Data Fig. 3j). PEN2 was originally identified as a component of γ-secretase 7 . Unlike PEN2, other subunits of γ-secretase did not directly participate in AMPK activation for low-dose metformin (Extended Data Figs. 3m-s, 4a, b and 6m-o; detailed discussions on the relationship between metformin and γ-secretase are provided in Supplementary Note 3). Imaging by confocal microscopy (Extended Data Fig. 4c), stochastic optical reconstruction microscopy (STORM; Fig. 2d) and APEX tag-based transmission electron microscopy ( Fig. 2e, with validation data in Extended Data Fig. 4d) showed that a portion of PEN2 (approximately 40%; Extended Data Fig. 4c) was localized on the lysosome. This finding was confirmed in subcellular fractionation assays (Extended Data Fig. 4e, with detailed discussions on PEN2 localization provided in Supplementary Note 4 and Extended Data Figs. 4f, g and 5a), and metformin did not alter the subcellular localization of PEN2 (Extended Data Fig. 5b, c). These results indicate that the pool of lysosomally localized PEN2 may have a distinct role, whereby it participates in metformin-induced AMPK activation (discussed in Supplementary Note 5). Indeed, constructs of PEN2 fused to other organelle-specific proteins did not restore AMPK activation by metformin when re-introduced into Pen2 -/-MEFs (Extended Data Fig. 5d, with validation data in Extended Data Fig. 5d, e).
High concentrations of metformin can increase cellular levels of AMP, which can allosterically activate AMPK; therefore, it was anticipated that AMPK activation induced by high metformin levels would be lysosome-independent. Indeed, high concentrations of metformin, which increased AMP/ATP and ADP/ATP ratios (Extended Data Fig. 1m-p), bypassed the requirement of PEN2 for AMPK activation, as did phenformin and buformin (Extended Data Fig. 5f). Moreover, PEN2 deficiency did not affect glucose-starvation-induced AMPK activation (Extended Data Fig. 5g) or other agonists (Extended Data Fig. 5g, h).
We next investigated the biophysical nature that underlies the binding of PEN2 to metformin. Differential scanning calorimetry assays showed a shift in the thermal transition midpoint in the presence of metformin (Extended Data Fig. 6a). Isothermal calorimetry (ITC) and surface plasmon resonance (SPR) measurements further gave estimated dissociation constant (K D ) values of 1.7 μM and 0.15 μM (with an association rate constant (k a ) value of 2,815 M -1 s −1 ), respectively. These values are within the range of detected intracellular metformin concentrations in animals or human patients administered with regular doses (Fig. 2f and Extended Data Fig. 6b, f, with detailed discussions in Supplementary Note 6). The ITC measurement gave an additional metformin binding site, with a much higher K D of 98 μM, which is beyond the ranges of clinically relevant intracellular concentrations of metformin (Extended Data Fig. 6b). As a control, other γ-secretase subunits did not show apparent binding affinity to metformin (Extended Data Fig. 6c). We also performed mass spectrometry on purified PEN2 conjugated to Met-P1 to identify the residue(s) responsible for binding metformin. As a result, the Y47 residue of PEN2 was identified (Extended Data Fig. 6d), which indicates that metformin may be able to bind the amino-terminal cytosolic face. In silico modelling further illustrated that at the N-terminal region of PEN2, metformin forms direct contacts with PEN2 through the F35 and E40 residues on PEN2 (Extended Data Fig. 6e). Indeed, mutation of both F35 and E40 to alanine on PEN2 (PEN2-2A) blocked its interaction with metformin ( Fig. 2g and Extended Data Fig. 6f). Re-introduction of PEN2-2A into Pen2 -/-MEFs did not restore metformin-induced AMPK activation, or v-ATPase inhibition, even though PEN2-2A shares a similar subcellular localization with wild-type PEN2 ( Fig. 2h and Extended Data Fig. 6h, with validation data in Extended Data Fig. 6g, i). The mass spectrometry results also revealed an additional, but much weaker, metformin-binding site at the carboxy-terminal (luminal) face of PEN2. Given that metformin may be transported through endocytosis and may be present in the lumen of lysosomes, we examined possible binding of metformin to the C terminus of PEN2. We found that mutation of residues at this site did not block metformin binding or dampen AMPK activation (Extended Data Fig. 6d, j-l).

ATP6AP1 tethers PEN2 to v-ATPase
We next investigated how metformin binding causes PEN2 to intersect with and inhibit v-ATPase. We analysed PEN2 that was immunoprecipitated after incubation with protein extracts of lysosomes by mass spectrometry. A total of 1,881 proteins were detected in the PEN2 prey, among which 889 were changed after metformin treatment. Of these 889 proteins, 123 are  Article lysosome-resident proteins (Supplementary Table 2). Among these 123 candidates, we were particularly interested in ATP6AP1 (also known as Ac45), an accessory factor of v-ATPase 8 , because its metformin-dependent interaction with PEN2 could be verified by co-immunoprecipitation assays in cells and in vitro (Fig. 3a-c and Extended Data Fig. 7a, b). Domain-mapping experiments identified that amino-acid residues from 420 to 440, which constitute the transmembrane domain of ATP6AP1, were responsible for PEN2 binding (Extended Data Fig. 7c). This finding was reinforced by results from experiments that used the chimeric construct LAMP2 TM -ATP6AP1, which has the ATP6AP1 transmembrane domain replaced by the transmembrane domain of the lysosomal protein LAMP2. This construct did not interact with PEN2 (Extended Data Fig. 7d).
In addition, PEN2 mutations on its interface towards ATP6AP1 (based on in silico docking assays; PEN2-20A), impaired the interaction between PEN2 a-c, Identification of ATP6AP1 as an interacting protein of PEN2. Lysates of HEK293T cells expressing HA-PEN2 or Myc-ATP6AP1 (a), and lysates from wild-type MEFs, Pen2 -/-MEFs (b) or Atp6ap1 -/-MEFs (c) were incubated with 10 μM metformin and immunoprecipitated (IP) for the PEN2 and AP1 proteins. d, Metformin does not promote the interaction between ATP6AP1 and PEN2-20A. HEK293T cells transfected with HA-tagged PEN2 or PEN2-20A were lysed and treated as in a. The interaction between ATP6AP1 and PEN2 was analysed by IP followed by IB. e-i, Loss of the PEN2-ATP6AP1 interaction abolishes the effects of metformin on AMPK activation. Atp6ap1 -/-MEFs re-introduced with ATP6AP1 Δ420-440 (e) or Pen2 -/-MEFs re-introduced with the PEN2-20A mutant (f) were treated with 200 μM metformin for 12 h followed by analysis of p-AMPK and p-ACC. g-i, The effects of ATP6AP1 and PEN2 mutants on the lysosomal translocation of AXIN (g), and the formation of the AXIN-based complex (h, i) were analysed. Concanamycin A (conA; 5 μM for 2 h) was used as a control. FL, full length. j, A schematic depicting that the metformin-PEN2-ATP6AP1 and the FBP-aldolase axes constitute two incoming shunts that converge at v-ATPase to elicit AMPK activation through the lysosomal pathway. For gel source data, see Supplementary Fig. 1. Data are the mean ± s.e.m., n values are labelled on each panel, and P values were calculated using two-sided Student's t-test (a, for Myc-ATP6AP1), two-sided Student's t-test with Welch's correction (a, for HA-PEN2) or two-way ANOVA, followed by Tukey's test (g). Experiments in this figure were performed three times, except for a (four times), and h and i (five times). and ATP6AP1 (Fig. 3d and Extended Data Fig. 7e-g). Of note, ATP6AP1 itself did not bind Met-P1 (Extended Data Fig. 7h). Together, these results indicate that after binding to metformin, lysosomal PEN2 is recruited to ATP6AP1 of the v-ATPase complex.
We next examined how ATP6AP1 mediates the inhibition of v-ATPase by metformin. First, as an integral member of v-ATPase, knockout of ATP6AP1 led to constitutive activation of AMPK (Extended Data Fig. 7ik). When we re-introduced the truncated ATP6AP1 mutant (Δ420-440), which lacks the transmembrane domain required for its interaction with PEN2, into Atp6ap1 -/-MEFs, the basal activity of v-ATPase was restored (Extended Data Fig. 8a, with validation data in Extended Data Fig. 7l). Of note, the ATP6AP1 Δ420-440 mutant did not mediate metformin-induced v-ATPase inhibition or AMPK activation ( Fig. 3e and Extended Data Fig. 8b, c). Similar restoration of v-ATPase activity, as well as blockade of AMPK activation, was observed when the LAMP2 TM -ATP6AP1 chimeric construct was re-introduced into Atp6ap1 -/-MEFs (Extended Data Fig. 8a, d, with validation data in Extended Data Fig. 7l). Furthermore, re-introduction of PEN2-20A, which cannot interact with ATP6AP1 even though it is localized in a similar manner as wild-type PEN2 (Extended Data Fig. 8e), into Pen2 -/-MEFs blocked the activation of AMPK and the inhibition of v-ATPase ( Fig. 3f and Extended Data Fig. 8f). These results indicate that metformin-bound PEN2, by gaining affinity to ATP6AP1, inhibits v-ATPase to activate AMPK.
We previously reported that glucose deprivation can activate lysosomal AMPK without increasing AMP/ADP levels through v-ATPase, Ragulator and AXIN 24 , which are downstream of the fructose-1,6-bisphosphate sensor aldolase. Knockout of AXIN, LAM-TOR1 (a subunit of Ragulator) or the v0c subunit of v-ATPase (ATP6v0c) in the liver, MEFs or HEK293T cells blocked the activation of AMPK by low-dose metformin (Extended Data Fig. 9a-h). Re-introduction of the AMPKβ1-G2A mutant, which cannot localize on lysosomes, into MEFs that are deficient in both AMPKβ1 and AMPKβ 2 also blocked the activation of AMPK by low-dose metformin (Extended Data Fig. 9a-h). Moreover, high concentrations of metformin bypassed the requirement for AXIN and LAMTOR1 in AMPK activation (Extended Data Fig. 9a- Supplementary Table 3). Ctrl, control. i, j, PEN2 and ATP6AP1 are required for AMPK activation induced by 0.1% metformin in the diet. The 5-week-old PEN2-LKO mice (i; tamoxifen was injected at 4 weeks old) or 8-week-old ATP6AP1-LKO mice expressing ATP6AP1 Δ420-440 ( j; viruses injected at 4 weeks old, and tamoxifen was injected at 5 weeks old), were fed with normal chow diet containing 0.1% metformin for 1 week, as previously described 10 .
Hepatic AMPK activation was then analysed by IB. For gel source data, see Supplementary Fig. 1. Data are shown as the mean ± s.e.m., n values are labelled on each panel, and P values were calculated using two-way repeated-measures ANOVA followed by Tukey's test (a, c and e compared blood glucose between the WT/ATP6AP1-FL + Met group and the PEN2-IKO/LKO/ATP6AP1 Δ420-440 + Met group at each time point; see also insets of a, c and e for area under the receiver operator characteristic curve (AUC) values, and P values by two-way ANOVA, followed by Tukey's test), two-sided Student's t-test (b, left), and two-way ANOVA, followed by Tukey's test (right panel of b, and d, f). Experiments in this figure were performed three times.

Article
LAMTOR1 through their regulation of v-ATPase. This is based on the fact that the lysosomal translocation of AXIN-and the formation of the AXIN-based complex-was dampened in Pen2 -/-MEFs, in Pen2 -/-MEFs expressing PEN2-2A or PEN2-20A mutants, and in Atp6ap1 -/-MEFs expressing the ATP6AP1 Δ420-440 mutant when treated with metformin ( Fig. 3g-i, Extended Data Figs. 9i-l and 10a, b). Blockade of v-ATPase by its inhibitor concanamycin A restored these phenotypes ( Fig. 3g-i, Extended Data Figs. 9i, k, 10a, b and 11a, b). As additional controls, aldolase and TRPV, which are required for signalling of low glucose to v-ATPase and AMPK 6,25 , were dispensable for the PEN2-sensed AMPK activation by metformin. This result was supported by the following lines of evidence: (1) expression of ALDOA-D34S, which mimics a high glucose state and blocks glucose-deprivation-induced AMPK activation in both mouse liver and cultured cells 6 (Extended Data Fig. 11c), did not block metformin-induced AMPK activation (Extended Data Fig. 11d, e); and (2) a quadruple knockout of Trpv1-Trpv4 in MEFs, or knockdown of Trpv2-Trpv4 in the liver of Trpv1 -/mice (leaving those cells or tissues with scarce TRPV expression 25 ), did not affect the activation of AMPK when treated with metformin (Extended Data Fig. 11f, g). Together, PEN2-ATP6AP1 relays the signal of metformin, as an intersecting shunt, to inhibit v-ATPase, which primes the lysosomal translocation of AXIN and LKB1 to the lysosomal surface for phosphorylation and activation of AMPK (schematically represented in Fig. 3j).

Phenotypes in animal models
We next explored the functions of PEN2 and ATP6AP1 to mediate the beneficial effects of metformin in animal models. We observed that mice that had PEN2 depleted specifically in the intestine (PEN2-IKO mice; generated as illustrated in Extended Data Fig. 12c, d), had impaired postprandial glucose-lowering effects of metformin, similar to those observed in intestine-specific Ampka knockout (AMPKα-IKO) mice (Extended Data Fig. 12a, b). We also observed impaired promotion of GLP-1 and insulin secretion by metformin (Fig. 4a, b and Extended Data Fig. 12e). Meanwhile, hepatic-specific depletion of PEN2 (PEN2-LKO mice; generated as illustrated in Extended Data Fig. 12f) led to strong impairments in the activation of AMPK in mouse liver. The effects of administration of metformin for 4 months to decrease levels of hepatic triglycerides (TAGs), as well as glucose tolerance in high-fat diet (HFD)-induced obese mice, were also impaired ( Fig. 4c, d and Extended Data Fig. 12g-k). Similarly, re-introduction of ATP6AP1 Δ420-440 into mouse liver with Atp6ap1 knocked out did not rescue the metformin effects on AMPK activation or on TAG level reduction (Fig. 4e, f and Extended Data Fig. 12l-r). Therefore, PEN2 and ATP6AP1 are required for the effect of metformin to reduce hepatic fat by activating the lysosomal AMPK pathway. We also tested whether lifespan extension induced by metformin depends on PEN2 and ATP6AP1. Consistent with previous reports 9 , metformin at 50 mM was able to extend the lifespan of C. elegans (Extended Data Fig. 13a), and no increases in AMP/ATP and ADP/ATP ratios were observed (Extended Data Fig. 13b). Knockdown of T28D6.9, the nematode orthologue of PEN2, blocked the metformin-induced lifespan extension effect in C. elegans (Fig. 4g, Extended Data Fig. 13b, c and Supplementary Table 3, with validation data in Extended Data Fig. 13d). Similarly, expression of mammalian ATP6AP1 Δ420-440 in ATP6AP1 -/-C. elegans impaired the metformin-induced AMPK activation and lifespan extension effects (Fig. 4h, Extended Data Fig. 13f, i and Supplementary Table 3). Of note, genetic manipulation of pen-2 and ATP6AP1 or the living bacteria (OP50 and HT115) on the culture plates did not affect the cellular uptake of metformin (Extended Data Fig. 13e, j, k). Finally, we examined the effects of normal chow diet that contained 0.1% metformin-a diet that has been shown to extend the lifespan and retard the ageing of mice through the activation of AMPK 10 -on the activation of AMPK in mice with hepatic depletion of PEN2 or expression of ATP6AP1 Δ420-440 . The activation of AMPK was strongly dampened in both of these mouse strains (Fig. 4i, j and Extended Data Fig. 13l). Taken together, PEN2, in conjunction with ATP6AP1, appears to be responsible for the three main biological benefits of metformin: lowering glucose levels, reducing hepatic fat content and extending lifespan.

Discussion
Here we identified that PEN2 is a target of metformin. After stimulation, PEN2 binds to the ATP6AP1 subunit of and inhibits the activity of v-ATPase without increasing AMP or ADP, which then activates lysosomal AMPK. The PEN2-ATP6AP1 axis therefore constitutes a signalling shunt that intersects the lysosomal v-ATPase-AXIN-AMPK axis, which enables metformin at low concentration to make use of the AMP-independent AMPK-activation pathway, which is can also be triggered by glucose starvation. We also established that the PEN2-ATP6AP1 pathway is not involved in AMPK activation at low glucose levels, which indicates that the PEN2-ATP6AP1 axis is a parallel route to the v-ATPase complex. Therefore, the two axes, PEN2-ATP6AP1 and aldolase-TRPV, sense the presence of metformin and the lowered levels of glucose, respectively, and impinge on v-ATPase to control the activation of AMPK. This finding underscores the important role of v-ATPase as a signalling node for lysosomal AMPK activation (Extended Data Fig. 13m).
We also showed that the PEN2-ATP6AP1 axis is required for the three main beneficial effects of metformin: postprandial glucose reduction, hepatic fat reduction and lifespan extension, all of which strictly depend on AMPK 9,20 (Extended Data Fig. 13g, h, 14a-i and Supplementary Notes 7-9). Our data are also consistent with previous findings that metformin can promote GLP-1 secretion in the intestine to lower blood glucose in an AMPK-dependent manner 21 , unless high doses of metformin are administered 26 . However, although it has been shown that metformin can also inhibit hepatic gluconeogenesis 1 , we found that low doses of metformin did not do so, as assessed by pyruvate tolerance tests and the quantification of gluconeogenic genes (Extended Data Fig. 14j, k). Moreover, we also found that the PEN2-ATP6AP1 axis is required for the inhibition of mTORC1 signalling by metformin (Extended Data Fig. 14l-n), in addition to the activation of AMPK. However, mTORC1 inhibition did not seem to be involved in the abovementioned beneficial effects mediated by AMPK (Supplementary Note 10).
The intersection of metformin signalling to the lysosomal AMPK pathway, without perturbing AMP/ADP levels, might underlie the reason why metformin exerts many benefits with few side effects. This pathway only allows the activation of a small pool of AMPK 6,27 , and this innate pathway is perhaps related to calorie restriction and would be less likely to cause adverse effects compared with global AMPK activation. It has been shown that indiscriminate AMPK activation results in harmful rather than beneficial effects. For example, the AMPK activator MK-8722, which appears to activate all AMPK subunit isoform combinations, can cause cardiac hypertrophy 28 . Moreover, naturally occurring mutations in PRKAG2 (encoding AMPKγ2), which cause increases in the basal activity of AMPK, are associated with cardiac disorders 29,30 . In summary, we identified that PEN2 is the molecular target for metformin and it intersect the glucose-sensing pathway to activate AMPK, which elicits benefits that resemble those induced by glucose starvation or calorie restriction. The PEN2-ATP6AP1 axis offers potential targets for screening for substitutes for metformin, which may be available to a wider range of tissues, such as muscle, thereby engendering better efficacy in treating diabetes and other metabolic diseases.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-022-04431-8.

Data reporting
The following sample sizes were similar to those previously used by us and others in this field: n = 4-6 human participants or mice were used to determine the pharmacokinetics of metformin 16,31,32 ; n = 5-7 mice were usually used to determine the effects of metformin on blood glucose 33,34 and fatty liver 20,35 ; n = 100-300 worms were used to determine lifespan 36,37 ; n = 20-42 cells from 2-6 dishes or experiments were included when conclusions were based on immunofluorescence staining 6,25 ; n = 3-5 samples were used for evaluation of the levels of AMP, ADP and ATP in cells and tissues 6,15,25,27 ; n = 3 samples to determine the expression levels and phosphorylation levels of a specific protein 24,38 ; n = 3 samples to determine the mRNA levels of a specific gene 20,24,39 ; and n = 3 samples to determine the activity of v-ATPase in vitro 25 . No statistical methods were used to predetermine sample sizes. All experimental findings were repeated as stated in the figure legends, and all additional replication attempts were successful. For animal experiments, mice or nematodes in each genotype were housed under the same condition or place. For cell experiments, cells of each genotype were cultured using the same condition. Each experiment was designed and performed along with proper controls, and samples for comparison were collected and analysed under the same conditions for the same batch of experiments. Randomization was applied wherever possible. For example, during MS analyses, samples were processed and subjected to the mass spectrometer in random order. For the animal experiments, sex-matched (only for mice), age-matched littermate mice in each genotype were randomly assigned to pharmacological or diet treatments. In cell experiments, cells of each genotype were parallel seeded and randomly assigned to different treatments. Otherwise, randomization was not performed. For example, when performing immunoblotting, samples needed to be loaded in a specific order to generate the final figures. Blinding was applied wherever possible. For example, samples, cages or agar plates during sample collection and processing were labelled as code names that were later revealed by the individual who picked and treated the animals or the cells, but did not participate in sample collection and processing until assessing the outcomes. Similarly, during microscopy data collection and statistical analyses, the fields of view were chosen on a random basis, and were often performed by different operators, thereby preventing potentially biased selection for desired phenotypes. Otherwise, blinding was not performed, such as the measurement of v-ATPase activity and the determination of metformin binding to PEN2 in vitro, for which operators had to know the conditions of each well and added reagents to the well accordingly during the measurement.

Determination of metformin pharmacokinetics in human participants
Women with obesity and diagnosed with polycystic ovary syndrome (PCOS) between September 2017 and July 2020 at Shanghai Jiao Tong University Affiliated Sixth People's Hospital were recruited. This study was approved by the Ethics Committee of Shanghai Sixth People's Hospital and was in accordance with the Declaration of Helsinki and Good Clinical Practice. All participants signed an informed consent form before enrolment. This study was registered on the Chinese Clinical Trial Registry (ChiCTR-IOR-17013169; http://www.chictr.org.cn/showproj.aspx?proj=21769).
Participants included in the study fulfilled the following criteria: (1) aged 18-40 years (inclusive); (2) body-mass index (BMI) higher than 27.5 kg m -2 ; and (3) meeting the diagnostic criteria of PCOS. The diagnostic criteria of PCOS were as follows: (1) irregular menstruation over the past year; (2) hyperandrogenism; and/or (3) ultrasound examination of polycystic ovaries. Participants with the following conditions were excluded: (1) after hysterectomy; (2) congenital adrenal hyperplasia, Cushing's syndrome or androgen-secreting tumours within 5 years; (3) mothers pregnant or lactating; (4) thrombosis-related history or risk factors, such as deep vein thrombosis, pulmonary embolism, myocardial infarction (angina), valvular heart disease, atrial fibrillation or cerebrovascular accident (transient ischaemic attack); (5) abnormal liver function (for example, caused by viral hepatitis); (6) a history of liver malignancy or adenoma, or a history of genital or breast malignancies; (7) a history of severe or frequent migraine attacks; (8) renal insufficiency; and (9) other factors that may affect the efficacy of the drug or cause complications by the drug.

Mouse strains
The Pen2 F/F mouse strain was generated according to an androgenetic haploid embryonic stem cell (AG-haESC)-based, CRISPR-Cas9-mediated genomic editing strategy as previously described 40 , but with minor modifications. The AG-haESCs carrying deletions in the differentially DNA methylated regions (DKO-AG-haESCs) were a gift from J.-S. Li (Shanghai Institute of Biochemistry and Cell Biology, CAS). The single guide RNAs (sgRNAs) targeting Pen2, 5′-GTGGTACTACTGCGCACGCG-3′ and 5′-GAAAGAATGAGCGAACGCCT-3′ (at intron 1 and intron 2, respectively) were cloned into a pSpCas9(BB)-P2A-mCherry-puromycin-sgRNA vector that was modified from the pSpCas9(BB)-2A-GFP vector (Addgene, 48138). The editing efficiency of each sgRNA was examined by transfecting into L929 cells followed by sequencing the targeted genomic segments before experiments. The template plasmid was constructed by inserting a 0.6-kb genomic fragment (containing exon 1 between the two loxP sites), along with a 1-kb fragment as a 5′-homology arm and a 1-kb fragment as 3′-homology arm into a pBKS vector. Approximately 1 × 10 6 DKO-AG-haESCs in a 6-well dish were transfected with 2 μg per well of each sgRNA (0.1 μg μl -1 ) and 4 μg of template plasmids (0.2 μg μl -1 ) with Lipofectamine 2000 transfection reagent. At 24 h after transfection, cells were trypsinized, washed with PBS and stained with 5 mg ml -1 Hoechst for 5 min. Those within the haploid 1C peak (haploid cells at G1 phase), mCherry-positive cells, were sorted on a FACSAria III flow cytometer (BD). Approximately 7,000 sorted cells were then cultured in a well of a gelatin-coated 6-well dish in EmbryoMax DMEM medium containing 10% ES-FBS and non-essential amino acids. Genotypes of individual clones were verified by sequencing, and the loxP-flanked Pen2 clones were selected for generating diploid ESCs. The verified clones were first treated with 0.05 μg ml -1 demecolcine for 10 h, digested to single cells and then intracytoplasmically injected into the mature oocyte (cultured in M2 medium) derived from the F 1 generation of C57BL/6 × DBA2 female mice at 8-10 weeks old. After 24 h, 2-cell ESCs were transplanted into pseudopregnant ICR female mice (8-10 weeks old, >26 g), and the offspring carrying the loxP-flanked Pen2 allele were further outcrossed 6 times to C57BL/6 mice before experiments.
Atp6ap1 F/F mice were generated according to the traditional, homologous recombination method 41 . In brief, a targeting vector was generated by inserting a 1-kb genomic fragment (containing exon 3 and exon 4) and an FRT-flanked PGK-Neo-polyA sequence (as a positive selection marker), along with a 3-kb fragment as 5′-homology arm and a 3-kb fragment, followed by a MC1-TK-polyA sequence (as a negative selection marker), as a 3′-homology arm, into a pBKS vector. The targeting vector was then linearized and purified. A total of 10 μg of the linearized vector (1 μg μl -1 ) was electroporated into 1 × 10 10 JM8A3 ESCs, followed by selection with G418, and genotyped by Southern blotting. Atp6ap1 F/F chimeric mice were obtained by microinjecting loxP-flanked Atp6ap1 ESC clones into C57BL/6 blastocysts (10-15 ESCs to a blastocyst), then transplanted into a pseudopregnant ICR female mice. The PGK-Neo allele was removed by crossing chimeric mice with ROSA26-FLPe mice, and was further outcrossed six times to C57BL/6 mice before experiments.
The Pen2 F/F mice were then crossed with Alb-CreERT2 or Villin-CreERT2 mice to generate inducible liver-specific Pen2 knockout mice (PEN2-LKO) or inducible intestine-specific Pen2 knockout mice (PEN2-IKO). The Atp6ap1 F/F mice were crossed with Alb-CreERT2 to generate inducible liver-specific Atp6ap1 knockout mice (ATP6AP1-LKO). Ampka1/2 F/F mice were crossed with Villin-CreERT2 mice to generate inducible intestine-specific Ampka knockout mice (AMPKα-IKO). Pen2, Atp6ap1 and Ampka were deleted by intraperitoneally injecting mice with tamoxifen (dissolved in corn oil) at 200 mg kg -1 , 3 times a week. Knockout efficiencies were analysed 1 week after the last injection by western blotting. ATP6AP1-LKO mice expressing wild-type ATP6AP1 or ATP6AP1 Δ420-440 were generated by injection into the tail vein different adeno-associated viruses (AAVs) carrying indicated inserts before knockout of the endogenous ATP6AP1 by tamoxifen. Levels of the re-introduced ATP6AP1 proteins were analysed 4 weeks after virus injection.
To generate Tg-ALDOA and Tg-ALDOA-D34S mice strains, human ALDOA or human ALDOA-D34S was cloned into the pLiv-Le6 vector (containing the constitutive human APOE gene promoter and its hepatic control region) between Cla I and Xho I sites. Vectors were then linearized with Not I and Spe I restriction enzymes. A 6.12-kb fragment was recovered using a QIAquick Gel Extraction kit (28706, Qiagen), followed by removal of endotoxin using an EndoFree Plasmid Maxi kit (12362, Qiagen). Plasmids were diluted to 2.5 ng μl -1 , and 1 pl was injected intranuclearly into the male pronucleus in zygote at embryonic day 0.5 of C57BL/6 mice. Two-cell embryos were transplanted into the pseudopregnant ICR female mice. The positive F 0 offspring were identified by sequencing and crossed with C57BL/6 mice. The F 1 mice carrying the transgenic genomic were selected by PCR, and ALDOA or ALDOA-D34S expression was examined by immunoblotting. One verified F 1 mouse of each genotype was chosen to set up the transgenic mouse strain, which was then outcrossed six times to C57BL/6 mice before experiments.

Metformin treatment of mice
The protocols described below for all mouse experiments were approved by the Institutional Animal Care and the Animal Committee of Xiamen University (XMULAC20180028). Unless stated otherwise, mice were housed with free access to water and standard diet (65% carbohydrate, 11% fat, 24% protein) under specific pathogen-free conditions. The light was on from 8:00 to 20:00, with the temperature kept at 21-24 °C and humidity at 40-70%. Male littermate controls were used throughout the study. Metformin was supplied either in drinking water at desired concentrations or in standard diet at 0.1% (w/w) for 1 week. For creating the diabetic mouse model, mice were fed a HFD (60% calories from fat; D12492, Research Diets) for desired time periods starting at 4 weeks old.
The following ages of mice were used. (1) For isolating primary hepatocytes: normal-chow-diet-fed wild-type and loxP-flanked Ampka mice aged 4 weeks (Fig. 1a, b, Extended Data Figs. 1c, e-g, 13g, h and 14f-h, n); normal-chow-diet-fed PEN2-LKO mice aged 6 weeks (Fig. 2b, Extended Data Figs. 3c, j and 14l, j; into which tamoxifen was injected at 4 weeks old); HFD-fed wild-type mice aged 38 weeks (Extended Data Fig. 12i; fed with a HFD for 34 weeks starting from 4 weeks old); HFD-fed PEN2-LKO mice aged 38 weeks (Extended Data Fig. 12i; into which tamoxifen was injected at 35 weeks old after 31 weeks of HFD treatment starting from 4 weeks old); normal-chow-diet-fed ATP6AP1-LKO mice expressing full-length ATP6AP1 or ATP6AP1 Δ420-440 aged 8 weeks (Extended Data Fig. 14m, j; into which an AAV carrying ATP6AP1 was injected at 4 weeks old and tamoxifen at 5 weeks old); and HFD-fed ATP6AP1-LKO mice expressing full-length ATP6AP1 or ATP6AP1 Δ420-440 aged 38 weeks (Extended Data Fig. 12p; into which an AAV was injected at 34 weeks old and tamoxifen at 35 weeks old, after 34 weeks of HFD treatment starting from 4 weeks old). (2) For glucose tolerance tests (GTTs), insulin tolerance tests (ITTs) and measurements of metformin, GLP-1, insulin and TAG contents: wild-type mice aged 5 weeks (Extended Data Fig. 1h, l, m); PEN2-IKO and AMPKα1/2-IKO mice aged 6 weeks (Fig. 4a, b and Extended Data Fig. 12b, e; mice at 4 weeks old were injected with tamoxifen and were fed a HFD for 1 week, and then treated with metformin for another week); HFD-fed PEN2-LKO mice aged 54 weeks (Fig. 4c, d and Extended Data Fig. 12g, h; mice at 4 weeks old were fed a HFD for 31 weeks, and then injected with tamoxifen; at 38 weeks old, mice were treated with metformin for 16 weeks); HFD-fed ATP6AP1-LKO mice expressing full-length ATP6AP1 or ATP6AP1 Δ420-440 aged 54 weeks (Fig. 4e, f and Extended Data Fig. 12n, o; mice at 4 weeks old were fed a HFD for 30 weeks, and were injected with an AAV at 34 weeks old; at 35 weeks old, the mice were injected with tamoxifen and then treated with metformin for 16 weeks). (3) For pyruvate tolerance tests (PTTs): PEN2-LKO mice aged 6 weeks (Extended Data Fig. 14k; mice at 4 weeks old were injected with tamoxifen and then treated with metformin for 1 week starting from 5 weeks old). (4) For immunoblotting and measurement of adenylates: wild-type mice aged 5 weeks (Extended Data Figs. 1j, k, m, n and 13l; mice were treated with metformin for 1 week starting from 4 weeks old); PEN2-LKO mice aged 6 weeks ( Fig. 4i; mice at 4 weeks old were injected with tamoxifen and then treated with metformin for 1 week starting from 5 weeks old); ATP6AP1-LKO mice expressing full-length ATP6AP1 or ATP6AP1 Δ420-440 aged 9 weeks ( Fig. 4j; mice at 4 weeks old were injected with an AAV; at 5 weeks old, the mice were injected with tamoxifen and then treated with metformin for 1 week starting from 8 weeks old); HFD-fed PEN2-LKO mice aged 39 weeks (Extended Data Fig. 12j; mice at 4 weeks old were fed a HFD for 31 weeks and then injected with tamoxifen; at 38 weeks old, mice were treated with metformin for 1 week); HFD-fed ATP6AP1-LKO mice expressing full-length ATP6AP1 or ATP6AP1 Δ420-440 aged 39 weeks (Extended Data Fig. 12q; mice at 4 weeks old were fed a HFD for 30 weeks and then injected with an AAV at 34 weeks old; at 35 weeks old, the mice were injected with tamoxifen and then treated with metformin for 1 week starting from 38 weeks old); AXIN-LKO mice, LAMTOR1-LKO mice and Tg-ALDOA-D34S aged 7 weeks (Extended Data Figs. 9d, e and 11e; mice were treated with metformin for 1 week starting from 6 weeks old); Tg-ALDOA-D34S aged 6 weeks (Extended Data Fig. 11c; mice were starved for 16 h); hepatic ATP6v0c knockdown mice aged 7 weeks (Extended Data Fig. 9f; mice were treated with metformin 1 week starting from 6 weeks old, into which an AAV carrying a short interfering siRNA (siRNA) against ATP6v0c was intravenously injected at 4 weeks old); and TRPV1 -/and hepatic TRPV2-TRPV4 knockdown mice aged 8 weeks (Extended Data Fig. 11g; mice were treated with metformin for 1 week starting from 7 weeks old, into which an AAV carrying siRNAs against TRPV-TRPV4 was intravenously injected at 5 weeks old). (5) For hepatic haematoxylin and eosin (H&E) staining: HFD-fed PEN2-LKO mice aged 54 weeks (Extended Data Fig. 12k; mice at 4 weeks old were fed a HFD for 31 weeks and then injected with tamoxifen; at 38 weeks old, mice were treated with metformin for 16 weeks); HFD-fed ATP6AP1-LKO mice expressing full-length ATP6AP1 or ATP6AP1 Δ420-440 aged 54 weeks (Extended Data Fig. 12r; mice at 4 weeks old were fed a HFD for 30 weeks and then injected with an AAV at 34 weeks old; at 35 weeks old, the mice were injected with tamoxifen and then treated with metformin for 16 weeks). (6) For all the other experiments, mice aged 4 weeks were used.

Serology, GTTs, ITTs and PTTs
Mice were individually caged for 1 week before each experiment. For GTTs, mice were fasted for 6 h (8:00 to 14:00), then gavaged or intraperitoneally injected with glucose at 1.5 g kg -1 (for lean mice) or 1 g kg -1 (for HFD-induced diabetic mice). Blood glucose was measured at the indicated time points through tail-vein bleeding using a OneTouch UltraVue automatic glucometer (LifeScan). ITTs were performed as per the GTTs, except that 1 U kg -1 insulin was intraperitoneally injected. PTTs were performed as GTTs, except that 1 g kg -1 sodium pyruvate was intraperitoneally injected into 16-h fasted (18:00 previous day to 10:00) mice. GTTs, ITTs and PTTs were performed using different batches of mice.
For measuring GLP-1 levels, 300 μl of blood from each mouse was collected in an ice-cold, K 2 EDTA spray-coated tube (366420, BD P800 blood collection system) containing pre-added 50 μl aprotinin (5 mg ml -1 ) and 50 μl diprotin A (5 mg ml -1 ) as previously described 42,43 (with modifications of concentrations of aprotinin and diprotin A used). Plasma was then prepared by centrifugation at 3,000g for 10 min at 4 °C, and 100 μl was used to determine the levels of GLP-1 using a GLP-1 EIA kit according to the manufacturer's instructions.
For measuring insulin levels, approximately 100 μl of mouse blood was collected at each time point (from the submandibular vein plexus) and was placed at room temperature for 20 min, followed by centrifugation at 3,000g for 10 min at 4 °C. A total of 5 μl of serum (the supernatant) was used to determine the levels of insulin using a Mouse Ultrasensitive Insulin ELISA kit according to the manufacturer's instructions. The five-parameter logistic fitted standard curve for calculating the concentration of insulin was generated from the Arigo Biolaboratories website (https://www.arigobio.cn/ELISA-calculator).

Histology
For H&E staining, liver tissues were fixed in 4% (v/v) paraformaldehyde for 24 h at room temperature then transferred to embedding cassettes. The cassettes were then washed in running water for 12 h, followed by successive soaking, each for 1 h, in 70% ethanol (v/v in water), 80% ethanol and 95% ethanol. The fixed tissues were further dehydrated in anhydrous ethanol for 1 h twice, followed by immersing in 50% xylene (v/v in ethanol) for 30 min, with two changes of xylene (15 min each) and two changes of paraffin wax (58-60 °C; 1 h each). The dehydrated tissues were embedded in paraffin on a HistoCore Arcadia paraffin embedding machine (Leica). Paraffin blocks were then sectioned at a thickness of 3 μm, dried on an adhesion microscope slide, followed by rehydrating in the following order: two changes of xylene at 70 °C 10 min each; two changes of anhydrous ethanol 5 min each; two changes of 95% ethanol 5 min each; one change of 80% ethanol for 5 min; one change of 70% ethanol for 5 min; one change of 50% ethanol for 5 min; and briefly in water. The sections were then stained in haematoxylin solution for 8 min, then washed in running water for 5 min, differentiated in 1% hydrochloric acid (in ethanol) for 30 s, washed in running water for 1 min, immersed in 0.2% (v/v in water) ammonium hydroxide solution for 30 s, washed in running water for 1 min and stained in eosin Y solution for 30 s. The stained sections were dehydrated in 70% ethanol for 5 min, twice in 95% ethanol for 5 min each, twice in anhydrous ethanol for 5 min each and two changes of xylene for 15 min each. The stained sections were mounted with Canada balsam and visualized on a Leica DM4 B microscope. Images were processed using LAS X software (v.3.0.2.16120, Leica), and formatted using Photoshop 2021 software (Adobe).
For measuring the TAG content, mice were euthanized, and the livers were immediately removed and rinsed in PBS for three times. Approximately 50 mg tissue was homogenized in 1 ml of PBS containing 5% (v/v) Triton X-100. The homogenates were boiled for 5 min followed by centrifugation at 20,000g at 25 °C for 10 min. The TAG content (from the supernatant) was determined using Labassay triglyceride reagent.
pen-2 was knocked down instead of knocked out in C. elegans because complete depletion of PEN-2 is lethal to C. elegans 7 . To knock down pen-2, the growth of nematodes was first synchronized: worms were washed off from agar plates with 15 ml M9 buffer (22.1 mM KH 2 PO 4 , 46.9 mM Na 2 HPO 4 , 85.5 mM NaCl and 1 mM MgSO 4 ) supplemented with 0.05% (v/v) Triton X-100 per plate, followed by centrifugation at 1,000g for 2 min. The worm sediment was suspended with 6 ml of M9 buffer containing 50% synchronizing bleaching solution (by mixing 25 ml of NaClO solution (5% active chlorine), 8.3 ml of 25% (w/v) NaOH and 66.7 ml of M9 buffer, for a total of 100 ml), followed by vigorous shaking for 2 min and centrifugation for 2 min at 1,000g. The sediment was washed with 12 ml of M9 buffer twice, then suspended with 6 ml of M9 buffer, followed by rotating at 20 °C, 30 r.p.m. for 12 h. Synchronized worms (around the L1 stage) were then placed on RNAi plates (NGM containing 1 mg ml -1 IPTG and 50 μg ml -1 carbenicillin) spread with HT115 E. coli stains containing RNAi against pen-2 (well A05 on plate 86 from C. elegans RNAi Collection (Ahringer)) for 2 days. The knockdown efficiency was examined by determining the levels of pen-2 mRNA by real-time quantitative PCR (qPCR). Approximately 1,000 worms were washed off a RNAi plate with 15 ml of M9 buffer containing Triton X-100, followed by centrifugation for 2 min at 1,000g. The sediment was then washed with 1 ml of M9 buffer twice, and then lysed with 1 ml of TRIzol. The worms were then frozen in liquid nitrogen, thawed at room temperature and then subjected to repeated freeze-thaw for another two times. The worm lysates were then placed at room temperature for 5 min, then mixed with 0.2 ml of chloroform followed by vigorous shaking for 15 s. After 3 min, lysates were centrifuged at 20,000g at 4 °C for 15 min, and 450 μl of the aqueous phase (upper layer) was transferred to a new RNase-free centrifuge tube, followed by mixing with 450 μl of isopropanol, then centrifuged at 20,000g at 4 °C for 10 min. The sediment was washed with 1 ml of 75% ethanol (v/v) followed by centrifugation at 20,000g for 10 min, and then with 1 ml of anhydrous ethanol followed by centrifugation at 20,000g for 10 min. The sediment was dissolved with 20 μl of RNase-free water after the ethanol was evaporated. The dissolved RNA was then reverse-transcribed to cDNA using ReverTra Ace qPCR RT master mix with a gDNA Remover kit, followed by performing real-time qPCR using Maxima SYBR Green/ROX qPCR master mix on a CFX96 thermocycler (Bio-Rad). Data were analysed using CFX Manager software (v.3.1, Bio-Rad). Knockdown efficiency was evaluated according to the CT value obtained. The primers for pen-2 are 5′-TACGTGATCGCCAGCATTGT-3′ and 5′-CGTGTGGACCGATTTCCTGA-3′. The primers for ama-1 (the internal control) are 5′-GACATTTGGCACTGCTTTGT-3′ and 5′-ACGATTGATTCCATGTCTCG-3′.
The ATP6AP1 -/-C. elegans strains expressing ATP6AP1 or its Δ420-440 mutant were established as follows: ATP6AP1 or its Δ420-440 mutant was introduced to the unc-76(e911) C. elegans strain, which had been outcrossed six times to the N2 strain; such generated strains were then subjected to knockout of the ATP6AP1 (vha-19) gene, and the uncoordinated phenotype of unc-76 was rescued. To generate unc-76 strains expressing ATP6AP1 or the Δ420-440 mutant, cDNA of ATP6AP1 or ATP6AP1 Δ420-440 was inserted into a pJM1 vector, with GFP as a selection marker, between the Nhe I and Kpn I sites (expressed under control by a sur-5 promoter), then injected into the syncytial gonad of the worm. Microinjection was performed using a Leica DMi8 microscope equipped with a M-152 manipulator (Narishige) and a microinjector system (Tritech). The injection pad was prepared by placing 2 drops (approximately 50 μl) of boiling 2% agarose (w/v) onto the centre of a glass coverslip (24 × 50 mm, 0.13-0.15 mm thickness), immediately followed by flattening with another coverslip, then dried at room temperature for 24 h. The injection needle was processed from a glass capillary (Borosil 1.0 × 0.75 mm ID/Fibre with Omega dot fibre, FHC) by a PC-100 Puller (Narishige) using the Step 2 programme (with heater no. 1 at 66 °C, and no. 2 at 75 °C), and was then loaded with 0.5 μl of pJM1-ATP6AP1 or ATP6AP1 Δ420-440 plasmid (200 ng μl -1 , centrifuged at 20,000g for 15 min before loading), and then connected to the nitrogen gas source at a pressure adjusted to 20 psi. Successful loading of plasmid was checked using a dissection microscope before loading onto the manipulator at a 45° angle in relation to the injection pad. All equipment was adjusted and aligned to make sure that a clear, centred view of the pad and the needle was obtained. The sealed needle tip was then opened by gentle tapping on the lateral side of the coverslip to a size that allowed a five-times broader droplet to be made during each injection. A young adult unc-76 worm with well distinguishable gonads was then gently anchored on an injection pad covered with a thin layer of microinjection oil (Halocarbon oil 700). The pJM1-ATP6AP1 or ATP6AP1 Δ420-440 plasmid was then injected into the syncytial arm of the gonad (gonadal sheath) until a slightly visible swelling was achieved. A drop of M9 buffer was then added to the injected worm, and the worm was floated on microinjection oil then picked and recovered on a NGM plate for 2 days. The F 1 GFP-expressing hermaphrodite was selected for further culture. The genomic sequence encoding ATP6AP1 (vha-19) was then knocked out from this strain by injecting a mixture of a pDD122 (Peft-3::Cas9 + ttTi5605 sgRNA) vector carrying sgRNAs against vha-19 (5′-CGTCGAAAAAA CCCGATTGTTGG-3′ for intron 2, and 5′-AATGATGTCAGGTTTTTTTCTGG-3′ for intron 3, designed using the CHOPCHOP website http://chopchop.cbu.uib. no/), and the p7616B (unc-76 (+)) rescue plasmid (100 μg ml -1 each) into young adult worms. The F 1 hermaphrodite worms with normal postures were individually cultured on a NGM plate. After egg-laying, worms were lysed using Single Worm lysis buffer (50 mM HEPES, pH 7.4, 1 mM EGTA, 1 mM MgCl 2 , 100 mM KCl, 10% (v/v) glycerol, 0.05% (v/v) NP-40, 0.5 mM DTT and protease inhibitor cocktail), followed by PCR with the primers 5′-AACTGCTTTTGGCTCGAAAATA-3′ and 5′-AAGTAAAAAGGGACAAAAGTCG-3′ for genotyping. The offspring generated from knockout-assured individuals were outcrossed six times to the N2 strain, and the expression levels of ATP6AP1 or ATP6AP1 Δ420-440 were examined by immunoblotting. Strains expressing ATP6AP1 or ATP6AP1 Δ420-440 at similar levels were chosen for further experiments.
The ages of the nematodes used in this study were as follows: (1) for lifespan assays, worms at L4 stage were used (Fig. 4g, h, Extended Data Figs. 13a and 14d, e; treated with metformin or N-acetylcysteine until death); (2) for analysis of adenylates and pharmacokinetics of metformin, p-AMPKα and reactive oxygen species (ROS) levels (Extended Data Figs. 13b, c, e, f, i-k and 14c, i), worms at L4 stage (after treatment of metformin for 1 day) were used; and (3) for the experiments using pen-2 knock down worms, worms at L1 stage were used (Extended Data Fig. 13d; fed with HT115 E. coli strain containing RNAi against pen-2).

Evaluation of nematode lifespan
Synchronized worms were cultured to the L4 stage before transfer to the desired agar plates. Worms were transferred to new plates every 2 days. Live and dead worms were counted during the transfer. Worms that displayed no movement after gentle touching with a platinum picker were judged as dead. Kaplan-Meier curves were generated using Prism 9 (GraphPad software), whereas the statistical analysis data were analysed using SPSS 27.0 (IBM).

Plasmids
Full-length cDNAs used in this study were obtained either by PCR using cDNA from MEFs or by purchasing from Origene, Sino Biological or Genescript. Mutations of PEN2 and ATP6AP1 were performed by PCR-based site-directed mutagenesis using PrimeSTAR HS polymerase (Takara). Expression plasmids for various epitope-tagged proteins were constructed in the pcDNA3.3 vector for transfection (ectopic expression) or in the pBOBI vector for lentivirus packaging (stable expression). To express Strep-tagged ATP6AP1, the ATP6AP1 cDNA was inserted into a modified pcDNA3.3-C-HA vector with the sequence encoding the HA epitope tag replaced by the sequence of the Strep tag 44 . PCR products were verified by sequencing (Invitrogen). The lentivirus-based vector pLV-H1-EF1a-puro was used for expression of siRNA in MEFs and HEK293T cells, and the AAV-based vector pAAV2 for mouse liver. The sequences for siRNAs of mouse Pen2 were as follows: 5′-GCCTGTGCCGGAAGTACTAT-3′ (1) and 5′-GTTCTTTGGTTAGTC AACATTT-3′ (2). All plasmids, except those used for adenovirus packaging (see below, the 'Packaging and injection of adenovirus and AAV' section), were purified using the caesium chloride density gradient ultracentrifugation method.

Primary hepatocytes
Human primary hepatocytes were isolated from surgically removed liver tissues. Fresh tissues were minced followed by digesting in 0.25% (w/v) trypsin supplemented with 0.5 mg ml -1 collagenase type IV for 10 min at 37 °C. Cells were then immediately plated (at 60-70% confluence) in collagen-coated 6-well plates in William's medium E plus 10% FBS, 100 IU penicillin and 100 mg ml -1 streptomycin. After 4 h of attachment, the medium was replaced with fresh William's medium E with 1% (w/v) BSA for another 12 h before further use. This study was approved by the Human Research Ethics Committee of the Shanghai Sixth People's hospital following the principles of the Declaration of Helsinki. Written informed consent was obtained from all participants.
Mouse primary hepatocytes were isolated with a modified two-step perfusion method using liver perfusion medium and liver digest buffer. Before isolation of hepatocytes, mice were first anaesthetized followed by the insertion od a 0.72 × 19 mm intravenous catheter into the postcava. After cutting off the portal vein, mice were perfused with 50 ml of liver perfusion medium at a rate of 5 ml min -1 , followed by 50 ml of liver digest buffer at a rate of 2.5 ml min -1 . The digested liver was then briefly rinsed with PBS and then dissected by gently tearing apart the Glisson's capsule with two sterilized, needle-pointed tweezers on a 6-cm dish containing 3 ml of PBS. The dispersed cells were mixed with 10 ml of ice-cold William's medium E plus 10% FBS and were filtered by passing through a 100-μm cell strainer (BD Falcon). Cells were then centrifuged at 50g at 4 °C for 2 min, followed by washing twice with 10 ml of ice-cold William's medium E plus 10% FBS. Cells were then plated and cultured as for human primary hepatocytes. Pen2 -/hepatocytes were established by infecting Pen2 F/F hepatocytes (isolated from loxP-flanked Pen2 mice) with adenoviruses expressing the Cre recombinase (or GFP as a control) for 6 h, followed by incubating in William's medium E with 1% (w/v) BSA for another 12 h before experiments.

Cell lines
HEK293T, AD293 (Adeno-X 293) cells, MEFs and L929 cells were maintained in Dulbecco's modified Eagle's medium supplemented with 10% FBS, 100 IU penicillin, 100 mg ml -1 streptomycin at 37 °C in a humidified incubator containing 5% CO 2 . The suspension HEK293T cell line, which was established from HEK293T cells, was a gift from Z. Wang (Xiamen Immocell Biotechnology). Cells were cultured in 400 ml of SMM 293-TII medium in a 2-litre conical glass flask at 37 °C, 160 r.p.m. on a CO 2 -resistant shaker (cat. 88881101, Thermo Scientific) in a humidified incubator containing 5% CO 2 . All cell lines were verified to be free of mycoplasma contamination. HEK293T cells were authenticated by STR sequencing. Polyethylenimine (PEI) at a final concentration of 10 μM was used to transfect HEK293T cells. Total DNA to be transfected for each plate was adjusted to the same amount by using a relevant empty vector. Transfected cells were collected 24 h after transfection.
Lentiviruses, including those for knockdown or stable expression, were packaged in HEK293T cells by transfection using Lipofectamine 2000. At 30 h after transfection, medium (approximately 2 ml) was collected and centrifuged at 5,000g for 3 min at room temperature. The supernatant was mixed with 10 μg ml -1 (final concentration) polybrene, and this was added to MEFs or HEK293T cells followed by centrifuging at 3,000g for 30 min at room temperature (spinfection). Cells were incubated for another 24 h (MEFs) or 12 h (HEK293T cells) before further treatments.
LAMTOR1 F/F , AXIN F/F and ATG5 F/F MEFs were established by introducing SV40 T antigen using lentivirus into cultured primary embryonic cells from mouse litters. LAMTOR1 -/-MEFs were generated by infecting LAMTOR1 F/F MEFs with adenoviruses expressing the Cre recombinase for 12 h, as for AXIN -/-MEFs, ATG5 -/-MEFs and APH1A-APH1C triple knockout MEFs. The infected cells were then incubated in fresh DMEM for another 12 h before further treatments. TRPV1-TRPV4 quadruple knockout MEFs were generated as previously described 25 .

Packaging and injection of adenovirus and AAV
Adenoviruses (AV) carrying Cre recombinase (Ad-Cre) were packaged using the AdEasy Adenoviral Vector system (240009, Agilent) in AD293 (Adeno-X 293) cells. In brief, pAdEasy vector carrying Cre recombinase was linearized with Pac I for 12 h, and the efficiency was confirmed by subjecting 0.2 μg of each linearized vector to 0.8% agarose gel (w/v, showing an approximately 30-kb band and a 4.5-kb band). The linearized vector was precipitated with two volumes of ethanol then dissolved with 20 μl of sterile water. A total of 5 μg of linearized vector was transfected into 3 × 10 6 AD293 cells cultured in a 60-mm dish by Lipofectamine 2000, and 3 ml of medium was refreshed after 12 h of transfection. Cells were cultured for another 7 days, with 1.5 ml of fresh medium added (not refreshed) every other day. Cells were collected together with the culture medium followed by three rounds of freeze-thaw cycles. Cell debris was removed by centrifugation at 20,000g for 10 min, and the supernatant was used to infect two 6-cm dishes of AD293 cells followed by 3 rounds of amplification to produce a 10-fold increase in titre. Viral particles were loaded on the top of 5 ml of 15% CsCl (dissolved in TBS (10 mM Tris, 0.9% (w/v) NaCl, 2.5% (w/v) sucrose, pH 8.1)) cushion over 4.5 ml of 40% CsCl cushion (w/v, dissolved in TBS) in an ultracentrifuge tube (cat. 344059, Beckman). The sample wase centrifuged at 30,000 r.p.m. in a SW41 rotor (Beckman) for 16 h at 4 °C. The heavier band was collected followed by dialysis in TBS for 1 h at 4 °C. Purified Ad-Cre viruses were stored at −80 °C.
AAVs were packaged in HEK293T cells using the protocol from Grieger et al. 45 . In brief, cells used for in-house viral production were maintained in 150-mm dishes. A total of 7 μg of pAAV-RC2/9 (AAV2 inverted terminal repeat (ITR) vectors pseudo-typed with AAV9 capsid) plasmid, 21 μg of pAAV-helper plasmid and 7 μg of pAAV2 plasmid (carrying ATP6AP1 or its mutant, or siRNAs against mouse Trpv2 to Trpv4) were added to 4 ml of DMEM without phenol red, followed by mixing with 175 μl of PEI solution (1 mg ml -1 , pH 7.5). The mixture was then incubated at room temperature for 20 min and then added to the dishes. At 60 h after transfection, cells were collected by scraping and centrifugation. The viral particles were purified from the pellet using an Optiprep gradient as previously described 45 . The titres of purified AAV were determined by real-time qPCR (see below). Viruses were stored at −80 °C before use, and were delivered to mice intravenously by lateral tail-vein injection. For each mouse, 1 × 10 11 particles of virus, adjusted to 200 μl of final volume (with PBS, pH 7.4), was injected.

IP and IB assays
For IP endogenous proteins, LAMTOR1, PEN2 and ATP6AP1 were immunoprecipitated and analysed as previously described 24 , but with minor modifications. In brief, ten 15-cm dishes of MEFs (grown to 80% confluence) were collected for IP of LAMTOR1, or two 10-cm dishes of MEFs (grown to 80% confluence) were collected for IP of PEN2 and ATP6AP1. Cells were lysed with 750 μl per dish of ice-cold ODG buffer (for LAM-TOR1; 50 mM Tris-HCl, pH 8.0, 50 mM NaCl, 1 mM EDTA, 2% (w/v) ODG, 5 mM β-mercaptoethanol with protease inhibitor cocktail) or lysis buffer (for PEN2 and ATP6AP1; 20 mM Tris-HCl, pH 7.5, 150 mM NaCl, 1 mM EDTA, 1 mM EGTA, 1% (v/w) Triton X-100, 2.5 mM sodium pyrophosphate, 1 mM β-glycerophosphate, with protease inhibitor cocktail), followed by sonication and centrifugation at 4 °C for 15 min. Cell lysates were incubated with respective antibodies overnight. Overnight protein aggregates were pre-cleared by centrifugation at 20,000g for 10 min, and protein A/G beads (1:250, balanced with ODG buffer or lysis buffer) were then added into the lysate-antibody mixture for another 3 h at 4 °C. The beads were centrifuged and washed with 100 times volume of ODG buffer or lysis buffer for 3 times (by centrifuging at 2,000g) at 4 °C and then mixed with an equal volume of 2× SDS sample buffer and boiled for 10 min before immunoblotting.
For IP of ectopically expressed PEN2 or ATP6AP1, a 6 cm-dish of HEK293T cells was transfected with different expression plasmids. At 24 h after transfection, cells were collected and lysed in 500 μl of ice-cold lysis buffer, followed by sonication and centrifugation at 4 °C for 15 min. Anti-HA (1:100) or anti-Myc (1:100) antibodies, along with protein A/G beads (1:100), or anti-FLAG M2 Affinity Gel (1:200, pre-balanced in lysis buffer) was added into the supernatant and mixed for 4 h at 4 °C. The beads were washed with 200 times volume of lysis buffer for 3 times at 4 °C and then mixed with an equal volume of 2× SDS sample buffer and boiled for 10 min before immunoblotting. Samples for probing APH1A and OCT1 were not boiled to avoid the formation of insoluble aggregates.
To analyse the levels of p-AMPKα and p-ACC in MEFs, HEK293T cells and primary hepatocytes, cells grown to 70-80% confluence (except for hepatocytes, which were grown to 60-70% confluence) in a well of a 6-well dish were lysed with 250 μl of ice-cold lysis buffer. The lysates were then centrifuged at 20,000g for 10 min at 4 °C and an equal volume of 2× SDS sample buffer was added into the supernatant. Samples were then boiled for 10 min and then directly subjected to IB.
To analyse the levels of p-AMPKα and p-ACC in liver, mice were anaesthetized after indicated treatments. Freshly excised (or freeze-clamped) tissue was lysed with ice-cold lysis buffer (10 μl mg -1 liver weight) followed by homogenization and centrifugation as described above. The lysates were then mixed with 2× SDS sample buffer and subjected to IB. To analyse the levels of p-AMPKα and p-ACC in nematodes, about 150 nematodes cultured on a NGM plate were collected for each sample. Worms were quickly washed with ice-cold M9 buffer containing Triton X-100, and were lysed with 150 μl of ice-cold lysis buffer. The lysates were then mixed with 5× SDS sample buffer, followed by homogenization and centrifugation as described above, and then subjected to IB. Analysis of PEN2 expression in mouse intestine was performed as previously described 46 . In brief, duodenal segments of intestine were removed immediately after euthanasia, and washed twice with pre-cold PBS containing inhibitor mix (1 mM PMSF, 5 μg ml -1 aprotinin, 1 μg ml -1 pepstatin A, 2 μg ml -1 leupeptin, 1 mM Na 3 VO 4 and 1 mM NaF). The tissues were then homogenized and sonicated in lysis buffer, followed by homogenization and centrifugation as described above. All samples, as described above, were subjected to IB on the same day of preparation, and any freeze-thaw cycle was avoided.
For IB, the SDS-polyacrylamide gels were prepared in house. In brief, the resolving gel solution (8%, 10 ml) was prepared by mixing 1.9 ml of 30% Acryl/Bis solution, 1 ml of 10× lower buffer (3.5 M Tris, 1% (w/v) SDS, pH 8.8), and 0.48 ml of 65% (w/v) sucrose (dissolved in water) with 6.62 ml of water; and the stacking gel solution (4%, 5 ml) was prepared by mixing 668 μl of 30% Acryl/Bis solution and 1.25 ml of 4× stacking buffer (0.5 M Tris, 0.4% (w/v) SDS, pH 6.8) with 3.08 ml water. For each glass gel plate (with 1.0-mm spacer, cat. 1653308 and 1653311, Bio-Rad), approximately 7 ml of resolving gel solution and 2.5 ml of stacking gel solution were required. APS (to 0.1% (w/v) final concentration) and TEMED (to 0.1% (v/v) final concentration) were added to the resolving gel solution. The resolving gel was overlaid with 2 ml of 75% (v/v) ethanol before acrylamide polymerization. After around 20 min (when a clear line between the resolving gel and ethanol is seen), the overlaid ethanol was poured off, dried using filter paper and then placed at room temperature for another 15 min to let the ethanol evaporate completely. The gel cassette was then filled with APS/TEMED-supplemented stacking gel solution, followed by placing a 15-well comb, and then placed at room temperature for 20 min. After removing the comb, the gel was rinsed with running buffer (25 mM Tris, 192 mM glycine, 1% (w/v) SDS, pH 8.3) before sample loading. Samples of less than 10 μl were loaded into wells, and the electrophoresis was run at 100 V by a Mini-PROTEAN Tetra Electrophoresis Cell (Bio-Rad). In this study, all samples were resolved on 8% resolving gels, except those for PEN2, APH1A-APH1C, PS1, PS2, LAMTOR1, cytochrome c and ATP6v0c, which were on 15% gels (prepared as those for 8%, except that a final concentration of 15% Acryl/Bis was added to the resolving gel solution), and AMPKβ1/2 and ALDOA, which were on 10% gels. To transfer the resolved proteins, the pre-cut PVDF membrane (0.45 μm, cat. IPVH00010, Merck) was incubated in methanol for 1 min, followed by equilibrating and soaking in pre-cooled transfer buffer (25 mM Tris, 192 mM glycine, 10% (v/v) methanol) for more than 5 min. After preparing the gel/membrane sandwich, the transfer was performed using a voltage at 100 V in a Mini Trans-Blot Cell (Bio-Rad) for 1 h at 4 °C. The blotted PVDF membrane was then incubated in blocking buffer (5% (w/v) BSA or 5% (w/v) non-fat milk (according to the instructions from the antibody suppliers) dissolved in TBST, which is composed of 40 mM Tris, 275 μM NaCl, 0.2% (v/v) Tween-20, pH 7.6) for another 2 h on an orbital shaker at room temperature, followed by rinsing with TBST twice, for 5 min each. The PVDF membrane was incubated with the desired primary antibody (diluted in TBST supplemented with 5% BSA and 0.01% (m/v) NaN 3 ) overnight at 4 °C on an orbital shaker with gentle shaking, followed by rinsing with TBST 3 times, 5 min each at room temperature, and then the secondary antibodies were incubated for 3 h at room temperature with gentle shaking. The secondary antibody (diluted in TBST) was then removed, and the PVDF membrane was further washed with TBST 3 times, 5 min each at room temperature. PVDF membranes were incubated in ECL mixture (by mixing equal volumes of ECL solution and peroxide solution for 5 min), then each membrane was placed onto a plastic wrap and laid with medical X-ray film (Fujifilm) in a light-proof cassette for the desired period of time. The films were then developed with X-OMAT MX developer and replenisher and X-OMAT MX fixer and replenisher solutions (Carestream) on a medical X-ray processor (Model 002, Carestream). The developed films were scanned using a Perfection V850 Pro scanner (Epson) using Epson scan software (v.3.9.3.4), and were cropped using Photoshop 2021 software (Adobe). Levels of total proteins and phosphorylated proteins were analysed on separate gels, and representative immunoblots are shown. The band intensities on developed films were quantified using Image J software (v.1.8.0, National Institutes of Health Freeware).

Quantification of G6pc1 and Pck1 mRNA levels by real-time PCR
Mouse primary hepatocytes cultured in 10-cm dishes were treated with metformin or with glucagon as a control. Total RNA was then prepared by lysing cells with 1 ml of TRIzol (for each 10-cm dish), followed by the addition of 270 μl of chloroform and vigorous mixing. After centrifugation at 12,000g for 15 min at 4 °C, 510 μl of the upper aqueous layer was transferred to a clean tube. The RNA was then precipitated by adding 675 μl of isopropanol followed by centrifugation at 12,000g for 15 min at 4 °C. The pellet was washed with 75% ethanol for 3 times by centrifugation at 12,000g for 5 min, and was dissolved in 30 μl of DEPC-treated water. The concentration of RNA was determined using a NanoDrop 2000 spectrophotometer (Thermo Scientific). A total of 4 μg of RNA was diluted with DEPC-treated water to a final volume of 10 μl at 65 °C for 5 min, and immediately chilled on ice. Random Primer Mix, Enzyme Mix and 5× RT buffer (all from the ReverTra Ace qPCR RT kit) were then added to the RNA solution, followed by incubation at 37 °C for 15 min, and then at 98 °C for 5 min on a thermocycler. The reverse-transcribed cDNA was quantified with Maxima SYBR Green/ROX qPCR master mix on a LightCycler 96 system (Roche) with the following programmes: pre-denaturing at 95 °C for 10 min; denaturing at 95 °C for 10 s, then annealing and extending at 60 °C for 30 s in each cycle; cycle number: 40. The following primer pairs were used for qPCR: mouse Actin, 5′-TTTGTGACCACAGCTGAGAGA-3′ and 5′-TGCCCATCAGGCAACTCG-3′; mouse G6pc1, 5′-TTGTGGCTTCCTTGGTCCTC-3′ and 5′-CAAAGG GAACTGTTGCGCTC-3′; mouse Pck1, 5′-CTCCTCAGCTGCATAACGGT-3′ and 5′-GTGGATA TACTCCGGCTGGC-3′. Data were analysed using Light-Cycler 96 software (v.1.1, Roche).

Identification of metformin-binding proteins
Lysosomes purified from sixty 10-cm dishes of MEFs (see below, 'Purification of lysosomes' section, for a detailed procedure) were lysed with 500 μl of ice-cold HEPES lysis buffer (50 mM HEPES, pH 8.0, 150 mM NaCl, 1% (v/v) NP40 with protease inhibitor cocktail), followed by sonication. After centrifugation, supernatants were incubated with 10 μM synthesized metformin probes (Met-Ps) at 4 °C for 2 h, then exposed to 365-nm wavelength UV (CX-2000, UVP) for 10 min. Supernatants were then adjusted to final concentrations of 1 mM TCEP, 0.1 mM TBTA, 1 mM CuSO 4 and 1 mM biotin-N 3 linker, and were incubated at 4 °C for 1 h. Protein aggregates were cleared by centrifugation at 20,000g for 15 min, and NeutrAvidin beads (1:100) were then added to the lysates for 2 h with gentle rotation. Beads were then washed with 100× volume of HEPES lysis buffer for 3 times at 4 °C, and then mixed with an equal volume of 2× SDS sample buffer, followed by SDS gel electrophoresis. The gels were stained with staining solution (1% (m/v) Coomassie Brilliant Blue R-250 dye dissolved in 45% (v/v) methanol and 10% (v/v) acetic acid in water) for 30 min, followed by decolouring with staining solution without R-250 dye before subjecting to MS analysis.
For validation of metformin binding by pull down, proteins were prepared from a 10-cm dish of HEK293T cells (grown to 80% confluence) transfected with 10 μg of indicated plasmids per dish for 24 h. Samples were similarly prepared as those using lysosomes, and were subjected to immunoblotting using the indicated antibodies.
For identification of metformin-binding sites on PEN2 by MS, 10 μg of FLAG-tagged PEN2 (expressed and purified from HEK293T cells) was used. Samples were similarly prepared as those using lysosomes except sonication.

Confocal microscopy
For determining the lysosomal localization of AXIN, cells grown to 80% confluence on coverslips in 6-well dishes were fixed for 20 min with 4% (v/v) formaldehyde in PBS at room temperature. The coverslips were rinsed twice with PBS and permeabilized with 0.1% (v/v) Triton X-100 in PBS for 5 min at 4 °C. After rinsing twice with PBS, the coverslips were incubated with anti-AXIN and anti-LAMP2 antibodies (both at 1:100, diluted in PBS) overnight at 4 °C. The cells were then rinsed 3 times with 1 ml of PBS, and then incubated with secondary antibody for 8 h at room temperature in the dark. Cells were washed for another 4 times with 1 ml of PBS, and then mounted on slides using ProLong Diamond antifade mountant. Confocal microscopy images were taken on a Zeiss laser scanning microscope (LSM) 780 with a ×63, 1.4 NA oil objective.
For detecting the pH of lysosomes, cells were grown on 35-mm glass-bottom dishes, and were cultured to 60-80% confluence. Cells were treated with 1 μM (final concentration) LysoSensor Green DND-189 for 1 h, then washed twice with PBS and incubated in fresh medium for another 30 min. In the meantime, 2 μg ml -1 (final concentration) Hoechst, along with ProLong Live antifade reagent, was added into the medium for staining the nucleus before taking images. For determining the mitochondrial membrane potential, cells were incubated with 10 μM JC-1 dye for 20 min at 37 °C, washed twice with PBS and then incubated in fresh medium before taking images. Mitochondrial membrane potentials were analysed by the red (emitted at 590 nm):green (emitted at 529 nm) fluorescence intensity ratio of JC-1 dye (excited at 488 nm). For determining ROS in MEFs, cells were incubated with 5 μM CellROX Deep Red for 30 min at 37 °C, washed twice with PBS and then incubated in fresh medium before taking images. During imaging, live cells were kept at 37 °C, 5% CO 2 in a humidified incubation chamber (Zeiss, Incubator PM S1). Images were taken using a Zeiss LSM 780 with a ×63, 1.4 NA oil objective.
For detecting ROS in C. elegans, synchronized worms cultured to L4 stage were treated with metformin for 24 h. Approximately 50 worms were soaked in 100 μl of M9 buffer containing 0.05% Triton X-100 and 10 μM CM-H 2 DCFDA for 30 min at 20 °C, followed by washing with M9 buffer containing 0.05% Triton X-100 three times. Worms were then placed on an injection pad prepared as described in the 'C. elegans strains' section, except that the 4% (w/v) agarose was used. Images were taken with a Zeiss LSM 900 with a ×20, 0.8 NA plan-Apochromat air objective.
For imaging PEN2, the Semi-intact IF protocol 47 was used. MEFs were grown on a 35 mm dish (cat. D35-20-10-N, In Vitro Scientific) to 50-60% confluence. Cells were rinsed with PBS once, and treated with buffer I (25 mM HEPES, pH 7.2, 125 mM potassium acetate, 5 mM magnesium acetate, 1 mM DTT, 1 mg l -1 glucose and 25 μg ml -1 digitonin) for 2 min on ice, and then buffer II (25 mM HEPES, pH 7.2, 125 mM potassium acetate, 5 mM magnesium acetate, 1 mM DTT and 1 mg l -1 glucose) for another 15 min on ice. The cells were then treated with 4% (v/v) formaldehyde in PBS at room temperature for 10 min. The slides were rinsed twice with PBS and then permeabilized with 0.05% (v/v) Triton X-100 in PBS for 5 min at 4 °C. After rinsing twice with PBS, the slides were blocked in block buffer (10% FBS (v/v) in PBS, with 0.1% (m/v) saponin) for 30 min. The slides were washed twice with PBS and incubated with primary antibodies diluted in block buffer overnight at 4 °C. The cells were then rinsed three times with PBS, and then incubated with a secondary antibody for another 8 h at 4 °C in the dark, followed by washing for four times with PBS and then mounted on slides using ProLong Diamond antifade mountant. Images were taken on a Zeiss LSM 780 with a ×63, 1.4 NA oil objective.
For taking images using the Zeiss LSM 780 system, the following lasers were used: Hoechst dye was excited with a Diode laser set at 405 nm; Lysosensor, JC-1, Alexa 488 and CM-H 2 DCFDA were visualized with an Ar gas laser (laser module LGK 7812) at 488 nm; Alexa 594 was visualized with a HeNe gas laser (LGK 7512 PF) at 594 nm; and CellROX Deep Red with a HeNe gas laser (LGK 7634) at 633 nm. When images were taken using the Zeiss LSM 900, CM-H 2 DCFDA was excited with laser module URGB (cat. 400102-9301-000, Toptica) using a 10-mW laser line. The parameters, including 'PMT voltage', 'Offset', 'Pinhole' and 'Gain', were kept unchanged between each picture taken. The resolution of images was 1,024 × 1,024 pixels. Images were processed using Zen 2012 software (for Zeiss LSM 780) or Zen 3.1 (for Zeiss LSM 900), and formatted using Photoshop 2021 software (Adobe). The intensities of CM-H 2 DCFDA in nematode intestines were quantified using ImageJ software (v.1.8.0, National Institutes of Health Freeware).

STORM imaging
MEFs stably expressing HA-PEN2 were cultured in a Lab-Tek II chambered no. 1.5 German coverglass system (155409, 8 Chamber, Nunc) to 50% confluence, and were treated following the Semi-intact IF protocol as described above, except that the cells were incubated in rabbit anti-HA tag primary antibody and rat anti-LAMP2 primary antibody, and then with the Atto 488 goat anti-rabbit IgG and Alexa-Fluor 647 donkey anti-mouse IgG secondary antibodies. The slides were then fixed with 4% (v/v) formaldehyde for another 10 min, and washed twice with PBS. The STORM imaging buffer with MEA was then prepared according to the manufacturer's instructions. In brief, 7 μl of GLOX (14 mg of glucose oxidase, 50 μl catalase (17 mg ml -1 ), 200 μl buffer A (10 mM Tris, pH 8.0 and 50 mM NaCl), vortexed to dissolve and cooled on ice) and 70 μl 1 M MEA (77 mg MEA dissolved in 1.0 ml of 0.25 M HCl) were added to 620 μl buffer B (50 mM Tris, pH 8.0, 10 mM NaCl and 10% (m/v) glucose) in a 1.5-ml Eppendorf tube, followed by a brief vortex. The mixture was then added to each well, and images were taken on an N-STORM (Nikon). The imaging was performed using an inverted microscope system (Ti-E Perfect Focus; Nikon) equipped with a monolithic laser combiner (MLC400, Agilent) containing solid-state lasers of wavelengths 405 nm, 488 nm and 561 nm at 50 mW (maximum fibre output power) and a 647-nm laser at 125 mW. After locating a suitable field, a diffraction-limited TIRF image was acquired for reference, followed by a STORM acquisition. The 647-nm laser was then sequentially fired at 100% power to excite all possible fluorophore molecules and photoswitch them into a non-emitting dark state, and then the 488-nm laser. The emitted wavelengths from Alexa Fluor 647 and Atto 488 fluorophores were then sequentially collected by the plan-Apochromat ×100/1.49 TIRF objective (Nikon), filtered by an emission filter set (Nikon TIRF Cube consisting of a TRF89902-EM filter set, Chroma Technology), and detected on an electron-multiplying charge-coupled device camera (Ixon DU-897, Andor Technology). During imaging, 20,000 sequential frames of each channel were acquired. The image acquisition, lateral drift correction and data processing were performed using NIS Elements software with STORM package (v.4.30 build 1053, Nikon) as previously described 48,49 .

Measurement of oxygen consumption rates
Cells were plated on a 96-well Seahorse XF cell culture microplate (Agilent). For primary hepatocytes, the microplates were pre-coated with collagen. Primary hepatocytes were plated at a density of 1,000 cells per well, whereas MEFs and HEK293T cells were plated at 10,000 cells per well. Cells were incubated in full medium (Williams' medium E containing 1% (w/v) BSA for hepatocytes, and DMEM containing 10% FBS for MEFs and HEK293T cells) overnight before experiments. Medium was then changed to Seahorse XF base medium (Agilent) supplemented with glucose (25 mM for MEFs and HEK293T cells, and 10 mM for hepatocytes), glutamine (4 mM) and pyruvate (1 mM), 1 h before the experiment. Cells were then placed in a CO 2 -free, XF96 Extracellular Flux Analyzer Prep Station (Agilent) at 37 °C for 1 h. The oxygen consumption rate was then measured at 37 °C in an XF96 Extracellular Flux Analyzer (Agilent), with a Seahorse XFe96 sensor cartridge (Agilent) pre-equilibrated in Seahorse XF Calibrant solution (Seahorse Bioscience, Agilent) in a CO 2 -free incubator at 37 °C overnight. Respiratory chain inhibitors (10 μM oligomycin, 10 μM FCCP, and a mixture of 1 μM antimycin A and 1 μM rotenone; all final concentrations) were then sequentially added to cells during the assay. Data were collected using Wave 2.6.1 Desktop software (Agilent) and exported to Prism 9 (Graph-Pad) for further analysis according to the manufacturer's instructions.

APEX-based TEM imaging
APEX2 imaging was performed as previously described 50 , but with minor modifications. In brief, HEK293T cells stably expressing APEX2-tagged proteins were grown on a 3.5-cm dish containing 1 ml of DMEM to approximately 70% confluence. Cells were fixed by gently adding 1 ml of 4% (v/v) glutaraldehyde solution (diluted in 1× phosphate buffer (0.1 M Na 2 HPO 4 :0.1 M NaH 2 PO 4 = 81:19, in water, pH 7.4), freshly prepared) pre-warmed to 37 °C. Cells were incubated in glutaraldehyde solution for 10 min, then substituted to 1 ml of ice-cold 2% (v/v) glutaraldehyde solution (freshly prepared) on ice for another 1 h, followed by 0.67% (v/v) glutaraldehyde (freshly prepared) solution at 4 °C overnight. Cells were then washed with 1 ml of ice-cold 1× phosphate Buffer for 5 times, 2 min each, followed by the addition of 1 ml of ice-cold 20 mM glycine solution (in 1× phosphate buffer) for 10 min on ice, and were then washed with 1× phosphate buffer for 5 times, 2 min each. Cells were then incubated in 1 ml of freshly prepared 1× DAB solution (0.5 mg ml -1 ) supplemented with 10 mM H 2 O 2 (all dissolved in 1× phosphate buffer) for 40 min, followed by washing with 1× phosphate buffer for 5 times, 2 min each. Cells were then stained with 2% (w/v) OsO 4 solution in 1× phosphate buffer on ice for 30 min, followed by washing for 5 times, 2 min each with ice-cold water. Cells were then stained in ice-cold 2% (w/v) uranyl acetate solution overnight at 4 °C in the dark, and were then washed for 5 times, 2 min each, with ice-cold water. Dehydration was then performed by sequentially incubating cells in the following ice-cold solutions: 20, 50, 70, 90 and 100% (v/v) ethanol (in water), each for 5 min, followed by incubation in anhydrous ethanol at room temperature for another 5 min. Cells were then quickly submerged ethanol/Spon 812 resin (3:1) mixture at room temperature for a 45-min incubation, and then in ethanol/resin (1:1) mixture at room temperature for 2 h, followed by ethanol/resin (1:3) at room temperature for 2 h, and finally 100% resin at room temperature for two rounds: first round overnight, and next round for 4 h. Resin was then completely drained, and the cells were spread with a thin layer of resin (with a total volume of approximately 400 μl, and below the thickness of 1 mm), followed by baking at 60 °C in a hot-wind drying oven for 48 h. The embedded cells were then sectioned into 75 nm slices after cooling down to room temperature. Images were taken on an electron microscope (Hitachi, HT-7800).

Purification of lysosomes
Lysosomes were purified using a lysosome isolation kit according to the manufacturer's instructions, but with minor modifications. In brief, MEFs from sixty 10-cm dishes (60-80% confluence), or 200 mg of mouse livers, were collected by directly scrapping at room temperature, followed by centrifugation for 5 min at 500g at 37 °C. Cells were resuspended in 7 ml of 1× extraction buffer containing protease inhibitor cocktail at room temperature, and were dounced in a 7-ml Dounce homogenizer (Sigma, cat. P0610) for 120 strokes on ice followed by centrifugation for 10 min at 1,000g, 4 °C, yielding post-nuclear supernatant (PNS). The PNS was then centrifuged for 20 min at 20,000g and the pellet was suspended in 1× extraction buffer by gentle pipetting, generating the crude lysosomal fraction (CLF). The volume of CLF was adjusted to 2.4 ml and then equally divided into six 1.5-ml Eppendorf tubes (400 μl per tube). A volume of 253 μl of OptiPrep and 137 μl of 1× OptiPrep dilution buffer were added to each CLF, and mixed by gentle pipetting. The mixture is defined as the diluted OptiPrep fraction (DOF). Each DOF (0.8 ml) was loaded into an 11 × 60 mm centrifuge tube at the top of 27% (0.4 ml) and 22.5% (0.5 ml) OptiPrep solution cushions, and then overlaid with 16% (1 ml), 12% (0.9 ml) and 8% (0.3 ml) OptiPrep solutions. The tube was then centrifuged on an SW60 Ti rotor (Beckman) at 150,000g for 4 h at 4 °C, and the fraction at the top of 12% OptiPrep solution was collected as the CLF. The fraction was diluted with two volumes of PBS, followed by centrifugation at 20,000g for 20 min. The supernatant was then aspirated, and the sediment was the lysosome fraction.

Measurement of v-ATPase activity in vitro
For each assay, lysosomes purified from two 10-cm dishes of MEFs were used. ATP hydrolysis activity was measured using a coupled spectrophotometric method as previously described 51 , but with some modifications. In brief, lysosomes were suspended in ATPase assay buffer (50 mM NaCl, 30 mM KCl, 20 mM HEPES-NaOH, pH 7.0, 10% (v/v) glycerol, 1 mM MgCl 2 , 1.5 mM phosphoenolpyruvate, 0.35 mM NADH, 20 U ml -1 pyruvate kinase and 10 U ml -1 lactate dehydrogenase) with 5 μM concanamycin A (for calculating v-ATPase-specific ATP hydrolysis activity) or DMSO, and pre-warmed at 37 °C for 10 min. The assay was initiated by the addition of 5 mM ATP, and the OD 341 was continuously recorded by a SpectraMax M5 microplate reader.
ATP-dependent proton transport activity was assessed by measuring the initial velocity (early reaction periods during the assay, before levelling off due to depletion of substrate) of ATP-dependent fluorescent quenching of FITC-dextran, as previously described 52,53 . In brief, lysosomes were loaded with FITC-dextran by incubating cells in DMEM supplemented with 2 mg ml -1 FITC-dextran (final concentration) on ice for 5 min, then transferred to a 37 °C incubator for 30 min. Cells were washed with DMEM for 3 times and incubated with DMEM for another 30 min at 37 °C to allow transport of FITC-dextran to lysosomes. Cells were collected and lysosomes were purified as described above. The lysosomes were resuspended in assay buffer (125 mM KCl, 1 mM EDTA, 20 mM HEPES, pH 7.5, with KOH) and were balanced on ice for 1 h, then mixed with 5 μM concanamycin A (for calculating the v-ATPase-specific proton transport activity) or DMSO, then warmed at 37 °C for 10 min. Fluorescence of FITC was recorded with excitation at 490 nm and emission at 520 nm using a SpectraMax M5 microplate reader. The initial slope of fluorescence quenching was measured after the addition of 5 mM Mg-ATP (final concentration).

Purification of mitochondria
Mitochondria were purified as previously described 54 , but with minor modifications 27 . In brief, forty 10-cm dishes of metformin-treated MEFs (60-80% confluence) were collected by direct scrapping at room temperature, followed by centrifugation for 5 min at 500g at 37 °C. Cells were then resuspended in 20 ml of ice-cold IB cells -1 buffer (225 mM mannitol, 75 mM sucrose, 0.1 mM EGTA and 30 mM Tris-HCl, pH 7.4), and dounced for 100 strokes in a 40-ml Dounce homogenizer (Sigma, cat. D9188), followed by two centrifugation rounds of 5 min at 600g at 4 °C. The supernatants were then collected and centrifuged for 10 min at 7,000g at 4 °C. The pellets were then washed twice with 20 ml of ice-cold IB cells -2 buffer (225 mM mannitol, 75 mM sucrose and 30 mM Tris-HCl pH 7.4). The suspensions were centrifuged at 7,000g, and again at 10,000g, both for 10 min at 4 °C. The pellets were then resuspended in 2 ml of ice-cold MRB buffer (250 mM mannitol, 5 mM HEPES pH 7.4 and 0.5 mM EGTA), and were loaded on top of 10 ml of Percoll medium (225 mM mannitol, 25 mM HEPES pH 7.4, 1 mM EGTA and 30% Percoll (v/v)) in 14 × 89-mm centrifuge tubes (cat. 344059, Beckman). The tubes were then centrifuged on a SW41 rotor (Beckman) at 95,000g for 0.5 h at 4 °C, and the dense band located approximately at the bottom of each tube was collected. The collected fractions were diluted with 10 volumes of MRB buffer, followed by centrifugation at 6,300g for 10 min at 4 °C; the pellets were resuspended and washed with 2 ml of MRB buffer, followed with centrifugation at 6,300g for 10 min at 4 °C. The pellets contained pure mitochondria.

Purification of cytosol
Cytosol was purified as previously described 55 . In brief, ten 10-cm dishes of cells were homogenized in 800 μl of the homogenization buffer (HB) containing 250 mM sucrose, 3 mM imidazole, pH 7.4. Homogenates were then passed through a 22-gauge needle attached to a 1-ml syringe for 6 times, and were then centrifuged at 2,000g for 10 min to yield PNS. PNS samples were then loaded on to the top of 11 × 60-mm centrifuge tubes that had been sequentially loaded with 1 ml of 40.6% sucrose (dissolved in HB), 1 ml of 35% sucrose (dissolved in HB), and 1 ml of 25% sucrose (dissolved in HB). Tubes were then centrifuged on an SW60 Ti rotor (Beckman) at 35,000 r.p.m. for 1 h at 4 °C, and the top fractions (about 200 μl) were collected as cytosolic fraction.

Protein expression
FLAG-tagged PEN2, as well as Strep-tagged ATP6AP1, was expressed in suspension HEK293T cells. A total of 200 ml cells at 5 × 10 6 per ml (with viability higher than 90%, as determined by Trypan Blue staining) were transfected with 200 μg of each plasmid dissolved in in 2 ml of SMM 293-TII medium containing 1,500 μg of PEI. After 48 h, cells were collected by centrifugation at 2,000g, and then lysed with 100 ml of ice-cold lysis buffer. The lysates were then sonicated and centrifuged at 20,000g for another 30 min at 4 °C. Anti-FLAG M2 Affinity Gel (1:100, balanced in lysis buffer), or streptavidin agarose (1:100, balanced in lysis buffer) was added into the supernatant and mixed for 4 h at 4 °C. The beads were then washed with 200 times volume of lysis buffer for 3 times at 4 °C. Proteins were then eluted with 1 ml of FLAG peptide (400 μg ml -1 final concentration) or desthiobiotin (2.5 mM final concentration) for another 1 h at 4 °C. Eluent was further diluted with 15 ml of PBS buffer, then concentrated to 1 ml using an Amicon Ultra-15 centrifugal filter unit with Ultracel-10 regenerated cellulose membrane. Such a process for buffer exchanging was performed for another two times before experiments. Of note, experiments involving proteins expressed and purified described above should be performed on the same day after the purification, and any freeze-thaw cycle should be avoided.

Differential scanning calorimetry
Differential scanning calorimetry assays were performed on a VP-DSC (GE Healthcare). The VP-DSC was run on a mode without feedback, and 15 min of equilibration at 20 °C was performed before and between each scan. The scanning range was set from 20 to 80 °C, and heating rate at 90 °C h -1 . The instrument was pre-equilibrated by running for five heating-cooling cycles with both the sample cell and the reference cell loaded with PBS. The sample cell was then loaded with 380 μl of FLAG-tagged PEN2 protein at 1 mg ml -1 in 20 μM metformin (in PBS) or PBS, and curves of heat capacity (Cp) versus temperature were recorded. Data were collected using VPViewer 2000 software (v2.66.7, GE Healthcare), and were then corrected for PBS baselines and normalized for scan rate and protein concentration 56 using Origin 7 software (v.7.0552, OriginLab).

ITC
ITC was performed using a MicroCal iTC200 isothermal titration calorimeter (GE Healthcare), with the sample cell and the reference cell maintained at 25 °C. The instrument was pre-equilibrated before the experiment. A total of 280 μl of FLAG-tagged PEN2 (at 60 μM) and 40 μl of metformin (2 mM stock solution), all in PBS buffer, were loaded into the sample cell and the injector, respectively, and PBS to the reference cell. Metformin was then titrated stepwise (0.4 μl during 0.8 s for the first injection, and 2 μl during 4 s for the rest of 19 injections, all injected at constant velocity, and with an interval of 150 s between each injection) into the PEN2 solution, or PBS as a control. During titration, the sample cell was continuously stirred at 1,000 r.p.m. Data of heat changes acquired during metformin-PEN2 titration were collected using ITC200 software (v.1.26.1, GE Healthcare), and were subtracted by those of metformin-PBS, then analysed using Origin 7 software (v.7.0552, OriginLab), which fits a 'Two Set of Sites' model.

SPR
Experiments were performed in triplicate at 25 °C on a BIAcore T200 using CM5 sensor chips, and data were analysed using BIAcore T200 Evaluation software (GE Healthcare) following the manufacturer's instruction. In brief, a cell on the CM5 sensor chip was activated with a mixture of 200 μM 1-ethyl-3-(3-dimethylaminopropyl)carbodiimide (EDC) and 50 μM N-hydroxysuccinimide (NHS) at 10 μl min -1 for 10 min. A total of 20 μl of FLAG-tagged PEN2 or PEN2-2A protein (1 mg ml -1 ) purified as described above and adjusted to pH 4.0 (by mixing with 180 μl of 10 mM sodium acetate solution, pH 4.0) was then immobilized on the surface of the cell at 10 μl min -1 for 5 min for two repetitive runs. The cell was then blocked with 1 M ethanolamine (10 μl min -1 for 10 min). A neighbouring cell that served as a reference was similarly activated and blocked, except that PBS adjusted to pH 4.0 was used for immobilization. Both of the cells were then equilibrated with PBS. Metformin stock solution (2 mM) was diluted to a series of concentrations (6.25, 3.125, 1.5625, 0.78125 and 0.05 μM (all in PBS)), and was flowed at 30 μl min -1 for 150 s in each run. At the end of each flow, cells were regenerated for 5 min with 10 mM glycine-HCl (pH 2.0) solution at 10 μl min -1 . Data from the sample cell were collected using BIAcore T200 Control software (v. 2.0, GE Healthcare), and were subtracted by those from the reference cell. Association and dissociation constants were obtained by global fitting of the data to a 1:1 Langmuir binding model using BIAcore T200 Evaluation software (v.2.0, GE Healthcare). Data were exported to Origin 7 software (v.7.0552, OriginLab) for generating the final figures.

Synthesis of metformin probes
Compounds were purified using a preparative HPLC (Sail 1000, Welch Materials) equipped with an XB-C18 column (30 × 250 mm, 5 μm, Welch Materials). Mass spectra for compound characterization were collected using an Autopurification LC Prep system equipped with an ACQUITY QDa detector (Waters) with the ESI mode. High-resolution mass spectra were obtained on a Q Exactive Orbitrap mass spectrometer (Thermo Scientific). NMR spectra were measured on an Avance III 600 MHz NMR spectrometer (Bruker) using tetramethylsilane as the internal standard, and the chemical shift was reported in δ (ppm), multiplicities (s = singlet, d = doublet, t = triplet, q = quartet, p = pentet, m = multiplet and br = broad), integration and coupling constants (J in Hz). 1 H and 13 C chemical shifts are relative to the solvent: δ H 2.50 and δ C 39.5 for DMSO-d 6 .

Protein and peptide MS
To identify metformin-interacting proteins, the pulled down, biotinylated proteins were subjected to SDS-PAGE. After staining with Coomassie Brilliant Blue R-250 dye, gels were decoloured and the excised gel segments were subjected to in-gel trypsin digestion and dried. Samples were analysed on a nanoElute (Bruker) coupled to a timsTOF Pro (Bruker) equipped with a CaptiveSpray source, or a NanoLC 425 System (SCIEX) coupled to a TripleTOF 5600+ mass spectrometer (SCIEX). Peptides were dissolved in 10 μl 0.1% formic acid (v/v) and were loaded onto a homemade C18 column (35 cm × 75 μm, ID of 1.9 μm, 100 Å). Samples were then eluted for 60 min with linear gradients of 3-35% acetonitrile (v/v, in 0.1% formic acid) at a flow rate of 0.3 μl min -1 . MS data were acquired with a timsTOF Pro mass spectrometer (Bruker) operated in PASEF mode, and were analysed using Peaks Studio software (X + , Bioinformatics Solutions), or a TripleTOF 5600+ mass spectrometer, and were analysed using ProteinPilot software (v.5.0, SCIEX). The mouse UniProt Reference Proteome database was used during data analysis.
To determine the PEN2-interacting proteins, the HA-tagged PEN2 immunoprecipitants (immunoprecipitated from twenty 10-cm dishes of MEFs stably expressing HA-tagged PEN2) were subjected to SDS-PAGE, and were processed as described above. Data acquisition was performed as described above, except that an EASY-nLC 1200 System (Thermo Scientific) coupled to an Orbitrap Fusion Lumos Tribrid spectrometer (Thermo Scientific) equipped with an EASY-Spray Nanosource was used. Data were analysed using Proteome Discoverer (v.2.2, Thermo Scientific) against the mouse UniProt Reference Proteome database.
For identifying metformin-binding sites on PEN2, the Met-P1conjugated FLAG-tagged PEN2 (purified from suspension HEK293T cells) were subjected to SDS-PAGE, and were processed as described above. MS analysis was performed as that for timsTOF Pro, except that the following parameters of Peaks Studio X + software were set: (1) precursor ion mass tolerance: 15 ppm; (2) fragment ion mass tolerance (error tolerance): 0.05 Da; (3) tryptic enzyme specificity with two missed cleavages: allowed; (4) mode of monoisotopic precursor mass and fragment ion mass was chosen; (5) mode of a fixed modification of cysteine carbamidomethylation was chosen; and (6) variable modifications including N-acetylation of proteins, oxidation@M and 222.17@X.

Measurement of adenylates
ATP, ADP and AMP from cells or tissues were analysed by capillary electrophoresis-based MS as previously described 6 . In brief, each measurement required cells collected from a 10-cm dish (60-70% confluence) or 100 mg of liver tissue dissected by freeze clamp. For analysis of metabolites, cells were rinsed with 20 ml of 5% (m/v) mannitol solution (dissolved in water) and instantly frozen in liquid nitrogen. Cells were then lysed with 1 ml of methanol containing Internal Standards 1 (IS1 (Human Metabolome Technologies, H3304-1002, 1:200), used to standardize the metabolite intensity and to adjust the migration time), and were scraped from the dish. For analysis of metabolites in liver, mice were anaesthetized after indicated treatments. The tissue was excised by freeze-clamping, then ground in 1 ml of methanol with 50 μM IS1. The lysate was then mixed with 1 ml of chloroform and 400 μl of water by 20 s of vortexing. After centrifugation at 15,000g for 15 min at 4 °C, 450 μl of aqueous phase was collected and was then filtrated through a 5-kDa cut-off filter (Millipore, cat. UFC3LCCNB-HMT) by centrifuging at 12,000g for 3 h at 4 °C. In parallel, quality control (QC) samples were prepared by combining 100 μl of the aqueous phase from each sample and then similarly filtered. The filtered aqueous phase was then freeze-dried in a vacuum concentrator (Labconco, CentriVap Benchtop centrifugal vacuum concentrator, equipped with a CentriVap −84 °C Cold Trap and a Scroll vacuum pump) at 4 °C, and then dissolved in 100 μl of water containing Internal Standards 3 (IS3 (Human Metabolome Technologies, H3304-1104, 1:200), to adjust the migration time). A total of 20 μl of redissolved solution was then loaded into an injection vial with a conical insert for CE-QTOF MS (Agilent Technologies 7100, equipped with 6545 mass spectrometer) analysis. Data were collected using MassHunter LC/MS acquisition 10.1.48 (Agilent), and were processed using Qualitative Analysis B.06.00 (Agilent). Levels of AMP, ADP and ATP were measured using full scan mode with m/z values of 346.0558, 426.0221, and 505.9885, respectively. Note that a portion of ADP and ATP could lose one phosphate group during in-source fragmentation, thus leaving the same m/z ratios as AMP and ADP, and should be corrected according to their different retention times in the capillary. Therefore, the total amount of ADP is the sum of the latter peak of the m/z 346.0558 spectrogramme and the former peak of the m/z 426.0221 spectrogramme, and the same is applied for ATP.
To analyse ATP, ADP and AMP in nematodes, HPLC-MS was performed. In brief, 150 nematodes maintained on NGM or siRNA plates (with or without 50 mM metformin) for 48 h were washed with ice-cold M9 buffer containing Triton X-100. Bacteria were removed by quickly spinning down the slurry at 100g for 5 s. Nematodes were then instantly lysed in 1 ml of methanol, then mixed with 1 ml of chloroform and 400 μl of water (containing 4 μg ml -1 [U-13 C]-glutamine), followed by 20 s of vortexing. After centrifugation at 15,000g for another 15 min at 4 °C, 800 μl of aqueous phase was collected, lyophilized in a vacuum concentrator at 4 °C, and then dissolved in 30 μl of 50% (v/v, in water) acetonitrile. Measurement of AMP and ATP level was based on ref. 60 using a QTRAP MS (SCIEX, QTRAP 5500) interfaced with a UPLC system (SCIEX, ExionLC AD). A total of 2 μl of each sample was loaded onto a HILIC column (ZIC-pHILIC, 5 μm, 2.1 × 100 mm, PN: 1.50462.0001, Millipore). The mobile phase consisted of 15 mM ammonium acetate containing 3 ml l -1 ammonium hydroxide (>28%, v/v) in the LC-MS-grade water (mobile phase A) and LC-MS-grade 90% (v/v) acetonitrile in LC-MS-grade water (mobile phase B) run at a flow rate of 0.2 ml min -1 . AMP, ADP and ATP were separated with the following HPLC gradient elution programme: 95% B held for 2 min, then to 45% B in 13 min, held for 3 min, and then back to 95% B for 4 min. The mass spectrometer was run on a Turbo V ion source in negative mode with a spray voltage of −4,500 V, source temperature at 550 °C, gas no.1 at 50 psi, gas no.2 at 55 psi, and curtain gas at 40 psi. Metabolites were measured using the multiple reactions monitoring mode, and declustering potentials and collision energies were optimized using analytical standards. The following transitions were used for monitoring each compound: 505.9/158.9 and 505.9/408.0 for ATP; 425.9/133.9, 425.9/158.8 and 425.9/328.0 for ADP; 345.9/79.9, 345.9/96.9 and 345.9/133.9 for AMP; and 149.9/114 for [U-13 C]-glutamine. Data were collected using Analyst 1.7.1 software (SCIEX), and the relative amounts of metabolites were analysed using MultiQuant 3.0.3 software (SCIEX). Note that a portion of ADP and ATP could lose one or two phosphate groups during in-source fragmentation, thus leaving same m/z ratios as AMP and ADP, which was corrected according to their different retention times in the column.
For quantification of AMP, ADP and ATP in cells, tissues or nematodes, [U-13 C, 15 N]AMP, [U-13 C, 15 N]ADP and [U-13 C, 15 N]ATP dissolved in individual lysates were used to generate corresponding standard curves by plotting the ratios of detected labelled AMP, ADP or ATP (areas) to the products of IS1 and IS3 (for CE-MS), or [U-13 C]-glutamine (for HPLC-MS), against the added concentrations of labelled AMP, ADP or ATP. The amounts of AMP, ADP and ATP were estimated according to standard curves, and were then divided by protein wet weight. The protein wet weight of each sample was determined by Bradford assay after dissolving the naturally dried protein sediment with 0.2 M KOH at room temperature.

Determination of metformin concentration
To measure metformin concentrations in serum, 50 μl of serum collected from each mouse or human participant was mixed with 80% methanol (v/v) in water using buformin at 100 μg l -1 as an internal standard. The aqueous phase was then collected after centrifugation at 15,000g for 15 min at 4 °C. Cells, liver tissues or intestinal tissues were prepared as in CE-MS measurement of adenylates, except that liver and intestinal tissues (50 mg each) were collected from anaesthetized, blood-drained mice, and no ultrafiltration was required. In addition, cells or liver tissues were rinsed with PBS before homogenization. Nematode samples were prepared as in for the HPLC-MS measurement of adenylates. Measurement was performed on a QTRAP MS (SCIEX, QTRAP 6500+) connected to a UPLC system (SCIEX, ExionLC AD). A total of 2 μl of each sample was loaded onto a pHILIC column (ZIC-pHILIC, 5 μm, 2.1 × 100 mm, PN: 1.50462.0001, Millipore). The mobile phase consisted of 10 mM ammonium formate containing 0.1% formate (v/v) in the LC-MS-grade water (mobile phase A) and LC-MS-grade acetonitrile containing 0.1% (v/v) formate in LC-MS-grade water (mobile phase B) run at a flow rate of 0.3 ml min -1 . The HPLC gradient was as follows: 95% B held for 1 min, then to 40% B in 6 min, held for 1 min, then to 95% B within 7.5 min, and held for 2.5 min. The QTRAP mass spectrometer was run on a Turbo V ion source and running in negative mode run in a spray voltage of −5,500 V, with source temperature at 500 °C, gas no.1 at 50 psi, gas no.2 at 55 psi, and curtain gas at 40 psi. Compounds were measured using the multiple reactions monitoring mode, and declustering potentials and collision energies were optimized using analytical standards. The following transitions were used for monitoring each compound: 130/71 for metformin and 158.1/60 for buformin. A standard curve was generated in each experiment for quantification. Data were collected using Analyst 1.6.3 software (SCIEX), and the relative amounts of metabolites were analysed using MultiQuant 3.0.2 software (SCIEX). The average cell volume of MEFs was estimated to be 2,263 μm 3 , HEK293T cells 4,240 μm 3 and primary hepatocytes 17,062 μm 3 using Imaris 7.4.0 software (Bitplane) from the axial image stacks of CDFA-SE labelled cells taken using a Zeiss LSM780.

Determination of TAG synthesis
TAG synthesis rates were determined by analysing the contents of labelled TAG in cells treated with [U-13 C] glucose or [U-13 C] palmitic acid (PA). Glucose was dissolved in PBS, and PA was conjugated to BSA before use. To conjugate PA, 200 mg of PA was first dissolved in 20 ml of ethanol in a conical flask by stirring, followed by dropwise mixing with 156 μl of 5 M NaOH. The slurry was constantly stirred for 12 h, which leads to a complete evaporation of ethanol. The dried sediment was then dissolved with 10% fatty-acid-free BSA to a final concentration of 2 mM.
Primary hepatocytes were isolated and cultured in DMEM containing 1% BSA for 12 h before the experiment. Cells were then incubated in glucose-free DMEM supplemented with 25 mM [U-13 C] glucose and 1% BSA, or DMEM containing 100 μM [U-13 C] PA and 1% BSA for another 12 h. Cells on a 10-cm dish were rinsed with 25 ml of PBS twice, and instantly frozen in liquid nitrogen. Cells were then lysed with 1 ml of methanol containing TAG (15:0/15:0/15:0) as an internal standard, and were scraped from the dish. The lysate was then quickly mixed with 1 ml of chloroform and 400 μl of water by 20 s of vortexing. After centrifugation at 15,000g for 15 min at 4 °C, 700 μl of organic phase was collected, followed by lyophilization with nitrogen blow on a pressured gas blowing concentrator (MGS-2200, EYELA) at room temperature. Analysis of TAG was performed on a Shimadzu Prominence UPLC system (Nexera UHPLC LC-30A) interfaced with a TripleTOF 5600+ system (SCIEX) equipped with an ESI source. Lyophilized samples were dissolved in 20 μl of dichloromethane/methanol solution (2/1, v/v), and was diluted with 380 μl of methanol/isopropanol/H 2 O solution (65/30/5, v/v/v). The injection volume was 5 μl. TAGs were separated through a C8 column (2.1 × 100 mm with 1.7 μm particle size, cat. 186002878, Waters) with column temperature maintained at 55 °C. Mobile phases consisted of 10 mM ammonium formate in acetonitrile/H 2 O (60/40, v/v) (mobile phase A) and 10 mM ammonium formate in isopropanol/acetonitrile (90/10, v/v) (mobile phase B) and was run at a flow rate of 0.26 ml min -1 . The gradient was as follows: 32% B for 1.5 min, then increased to 97% B within 19.5 min and held for 4 min, then back to 32% B and held for another 5 min. The flow rate for mobile phases was set at 0.26 ml min -1 . The mass spectrometer was run in positive, information-dependent acquisition (IDA) mode, with the source temperature of 550 °C, the ion source gas 1 and 2 at 55 psi, the curtain gas at 35 psi, the collision energy at 40 eV, the ion spray voltage floating at 5.5 kV, and the mass range at 500-1,250 m/z. The accumulation time for full scan was set at 150 ms, and the accumulation time for each IDA scan was 45 ms. Peaks of metabolites with intensities larger than 100 c.p.s. after adding up the signal from 10 rounds of IDA scans were chosen for further analysis. Data were collected using Analyst TF 1.6 software (SCIEX), and were analysed using MS-DIAL 4.7 software (RIKEN), through which the deconvolution and streamline criteria were used for peak/TAG identification.

Determination of β-oxidation rates
The rates of β-oxidation were determined through the labelled intermediates of the TCA cycle in cells treated with [U-13 C] PA for a certain time duration. Primary hepatocytes were cultured and treated as those used for the determination of TAG synthesis. Cells were rinsed with PBS twice, froze in liquid nitrogen and then lysed with 1 ml of 80% methanol (v/v) in water containing 10 μg ml -1 myristic-d27 acid as an internal standard, followed by 20 s of vortexing. After centrifugation at 15,000g for 1 min at 4 °C, 600 μl of supernatant (aqueous phase) was freeze-dried at 4 °C. The lyophilized sample was then vortexed for 1 min after mixing with 50 μl of freshly prepared methoxyamine hydrochloride (20 mg ml -1 in pyridine), followed by incubating at 4 °C for 1 h. The mixture was sonicated at 0 °C by bathing in an ice slurry for 10 min, and was then incubated at 37 °C for 1.5 h, followed by mixing with 50 μl of MTBSTFA and incubated at 55 °C for 1 h. Before subjecting to GC-MS, samples were centrifuged at 15,000g for 10 min, and 60 μl of supernatant was loaded into an injection vial. GC was performed on a HP-5MS column (30 m × 0.25 mm i.d., 0.25 μm film thickness) using a GC/MSD instrument (7890-5977B, Agilent). The injector temperature was 260 °C. The column oven temperature was first held at 70 °C for 2 min, then increased to 180 °C at the rate of 7 °C min -1 , then to 250 °C at the rate of 5 °C min -1 , then to 310 °C at the rate of 25 °C min -1 , where it was held for 15 min. The MSD transfer temperature was 280 °C. The MS quadrupole and source temperature were maintained at 150 °C and 230 °C, respectively. Data were collected using MassHunter GC/ MS Acquisition software (B.07.04.2260, Agilent), and were analysed using GC-MS MassHunter Workstation Qualitative Analysis software (v.B.07.01SP1, Agilent).

Statistical and reproducibility
Statistical analyses were performed using Prism 9 (GraphPad software), except for the survival curves, which were analysed using SPSS 27.0 (IBM). Each group of data was subjected to Kolmogorov-Smirnov test, Anderson-Darling test, D'Agostino-Pearson omnibus test or Shapiro-Wilk test for normal distribution where applicable. Unpaired two-tailed Student's t-test was used to determine significance between two groups of normally distributed data. Welch's correction was used for groups with unequal variances. Unpaired two-tailed Mann-Whitney test was used to determine significance between data without a normal distribution. For comparisons between multiple groups, an ordinary one-way or two-way analysis of variance (ANOVA) was used, followed by Tukey's, Sidak's, Dunnett's or Dunn's test as specified in the figure legends. The assumptions of homogeneity of error variances were tested using F-test (P > 0.05). For comparison between multiple groups with two fixed factors, an ordinary two-way ANOVA or two-way repeated measures ANOVA (for GTT, ITT and PTT data) was used, followed by Tukey's or Sidak's multiple comparisons test as specified in the legends. Geisser-Greenhouse's correction was used where applicable. The adjusted means and s.e.m. or s.d. were recorded when the analysis met the above standards. Differences were considered significant when P < 0.05, or P > 0.05 with large differences of observed effects (as suggested in refs. 61,62 ). All specific statistical details can be found in the figure captions and source data. All images shown without biological replicates are representative of a minimum of three independent experiments.

Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this paper.

Data availability
The data supporting the findings of this study are available within the paper and its Supplementary Information files. The MS proteomics data have been deposited to the ProteomeXchange Consortium (http:// proteomecentral.proteomexchange.org) through the iProX partner repository 63 with the dataset identifier PXD030090. Materials, reagents or other experimental data are available upon request. Full immunoblots are provided as Supplementary Information Fig. 1. Source data are provided with this paper.

Article
Extended Data Fig. 13 | PEN2-ATP6AP1 axis is required for metforminmediated lifespan extension in nematodes. a, Metformin extends lifespan of C. elegans. Wildtype nematodes (N2) were cultured on NGM plates containing metformin at different concentrations. Lifespans were determined, and results are shown as Kaplan-Meier curves, see also statistical analysis data on Supplementary Table 3. b, d, Metformin at 50 mM does not elevate AMP/ADP levels in C. elegans with knockdown of PEN2. Nematodes were knocked down of PEN2 (b, generated as in Fig. 4g by culturing on RNAi plates containing E. coli expressing siRNA against PEN2; the knockdown efficiency was assessed by determining mRNA levels of PEN2 as shown in d; results are mean ± s.e.m., n = 3 for each genotype; and P value by two-sided Student's t-test), and were maintained on RNAi plates containing 50 mM metformin for 2 more days. Worms were then subjected to the analysis of AMP:ATP and ADP:ATP ratios. Data are shown as mean ± s.e.m., n = 4 for each genotype/treatment, and P value by two-way ANOVA, followed by Tukey. c, PEN2 is required for metformininduced lifespan extension in C. elegans. Nematodes were generated as in Fig. 4g and AMPK activation after 2-day treatment of metformin, were determined. e, j, Metformin uptake is not altered in nematodes lacking PEN2 or ATP6AP1. L4 larvae of nematodes with PEN2 knockdown (e, generated as in Fig. 4g), or of nematodes with knockout of ATP6AP1 and stable expression of full length ATP6AP1 or ATP6AP1 Δ420-440 mutant ( j, generated as in Fig. 4h), were cultured on RNAi plates (e) or NGM plates ( j) containing 50 mM metformin, for 2 days. Metformin concentrations were determined by HPLC-MS, and normalised to protein concentrations. Data are shown as mean ± s.e.m., n = 4 for each genotype; and P value by two-sided Student's t-test. f, Metformin at 50 mM does not elevate AMP/ADP levels in ATP6AP1 -/-C. elegans re-introduced with ATP6AP1 Δ420-440 . Nematodes were knocked out of ATP6AP1 (vha- 19), then reintroduced with ATP6AP1 Δ420-440 or ATP6AP1 as a control, as shown in Fig. 4h, and then cultured on NGM plates containing 50 mM metformin for 2 more days before subjecting to analysis of AMP:ATP and ADP:ATP ratios. Data are shown as mean ± s.e.m., n = 4 for each genotype/treatment; and P value by two-way ANOVA, followed by Tukey. g, Low metformin fails to promote fatty acid β-oxidation. AMPKα1/2 -/mouse primary hepatocytes were treated with 100 μM BSA-conjugated [U-13 C]-palmitic acid and 5 μM or 500 μM metformin for 4 h. Relative contents of labelled citrate, succinate and fumarate were determined by GC-MS. Data are shown as mean ± s.e.m., n = 4 for each genotype/treatment; and P value by two-way ANOVA, followed by Tukey. h, Low metformin inhibits de novo lipogenesis, but does not affect TAG synthesis from fatty acids. Mouse primary hepatocytes were isolated as in g, and were treated with 100 μM BSAconjugated [U-13 C]-palmitic acid (left panel) and 25 mM [U-13 C]-glucose (right panel) for 12 h for determining the TAG synthesis from fatty acids and de novo lipogenesis, respectively. Metformin at 5 μM or 500 μM was added to the culture medium along with labelled palmitic acid or glucose. Relative contents of labelled TAG were determined by LC-MS. Data are shown as mean ± s.e.m., n = 4 for the determination of TAG synthesis and n = 5 for the determination of de novo lipogenesis; and P value by two-way ANOVA, followed by Tukey. i, ATP6AP1 is required for metformin-induced lifespan extension in C. elegans. Nematodes were treated as in Fig. 4h, AMPK activation after 2-day treatment of metformin, were determined. k, The presence of living bacteria on the culture plates does not affect metformin uptake by nematodes. The L4 larvae of N2 were cultured on NGM plates (left panel) or RNAi plates (right panel) containing 50 mM metformin. Metformin concentrations of nematodes at day 1 and day 3 (before culture plate refreshing) were measured by HPLC-MS. Data are shown as mean ± s.e.m., n = 4 for each treatment, and P value by two-sided Student's t-test. l, No changes in AMP:ATP and ADP:ATP ratios in the liver from mice fed with 0.1% metformin in diet. Mice at 4 weeks old were fed with normal chow (NC) diet, or normal chow diet containing 0.1% metformin (This way of giving metformin has been shown to be effective in extending the lifespan of mice 10 ) for a week. Hepatic AMP:ATP and ADP:ATP ratios were then determined. Results are shown as mean ± s.e.m., n = 9 for normal chow or n = 7 for 0.1% metformin; and P value by two-sided Student's t-test. m, Schematic diagramme showing that the FBP-unoccupied aldolase-triggered glucose sensing and the PEN2-mediated metformin signalling axes converge at the v-ATPase complex to trigger AMPK activation. In response to lowering glucose levels, aldolase inhibits the cation channel TRPV, the latter of which then disrupts the association of the former with, and inhibits, v-ATPase. Metformin at low concentrations binds to PEN2, and metformin-bound PEN2 is recruited to v-ATPase via interacting with ATP6AP1, thereby inhibiting the activity of v-ATPase. Also shown is that high concentrations of metformin activates AMPK via elevating cellular AMP levels. Experiments in this figure were performed three times, except l five times.