Loss of hepatic AMP-activated protein kinase impedes the rate of glycogenolysis but not gluconeogenic fluxes in exercising mice

Pathologies including diabetes and conditions such as exercise place an unusual demand on liver energy metabolism, and this demand induces a state of energy discharge. Hepatic AMP-activated protein kinase (AMPK) has been proposed to inhibit anabolic processes such as gluconeogenesis in response to cellular energy stress. However, both AMPK activation and glucose release from the liver are increased during exercise. Here, we sought to test the role of hepatic AMPK in the regulation of in vivo glucose-producing and citric acid cycle–related fluxes during an acute bout of muscular work. We used 2H/13C metabolic flux analysis to quantify intermediary metabolism fluxes in both sedentary and treadmill-running mice. Additionally, liver-specific AMPK α1 and α2 subunit KO and WT mice were utilized. Exercise caused an increase in endogenous glucose production, glycogenolysis, and gluconeogenesis from phosphoenolpyruvate. Citric acid cycle fluxes, pyruvate cycling, anaplerosis, and cataplerosis were also elevated during this exercise. Sedentary nutrient fluxes in the postabsorptive state were comparable for the WT and KO mice. However, the increment in the endogenous rate of glucose appearance during exercise was blunted in the KO mice because of a diminished glycogenolytic flux. This lower rate of glycogenolysis was associated with lower hepatic glycogen content before the onset of exercise and prompted a reduction in arterial glucose during exercise. These results indicate that liver AMPKα1α2 is required for maintaining glucose homeostasis during an acute bout of exercise.

The dysregulation of glucose homeostasis is associated with adverse consequences and is of great concern in pathologies such as diabetes (1). In addition to disease states, physiological conditions such as exercise challenge the ability of the gluco-regulatory system to maintain blood glucose within a narrow range (2). Systemic glucose disposal is elevated during shortterm, moderate exercise without appreciably changing arterial glucose (3)(4)(5)(6). This is due to a concurrent increase in glucose release from the liver (5,7,8).
Hepatic glucose production during exercise is influenced by multiple factors, including endocrine action (9). In particular, glucagon increases hepatic glycogenolysis and gluconeogenesis (7). It also stimulates gluconeogenic precursor extraction by the liver (7,8). Of note, glucose synthesis from three-carbon sources is energetically demanding, making ATP of great importance for this anabolic process (10). Consequently, exercise promotes liver energetic stress, as indicated by an increase in the AMP/ATP ratio (11).
Within the cell, AMP-activated protein kinase (AMPK) 2 responds to changes in the AMP/ATP ratio (12). This heterotrimeric serine/threonine kinase is composed of a catalytic ␣ subunit and regulatory ␤ and ␥ subunits. Two catalytic subunit isoforms are present in the liver: ␣1 and ␣2 (13)(14)(15). Kinase activity of the ␣ subunits is enhanced by energetic stress and post-translational modification. AMP binds the AMPK ␥ subunit (16). This promotes a conformational change in the enzyme that positions itself for phosphorylation on threonine residue 172 of the ␣ subunit (␣Thr-172) and protects against ␣Thr-172 dephosphorylation (17)(18)(19). Additionally, ATP com-petitively impedes AMPK activation by AMP (20). In response to energy discharge, AMPK promotes oxidative metabolism to maintain energy status (13,21). For example, loss of hepatic AMPK attenuates oxygen consumption in vitro (22,23), suggesting that AMPK activity is critical for appropriate mitochondrial oxidative phosphorylation and ATP provision.
Experimental and physiological conditions with elevated glucagon action, including exercise, promote AMPK activation (11,24,25). This is consistent with the need for adequate ATP supply to support gluconeogenesis during muscular work. It is important to note that many hypotheses regarding the physiological influence of hepatic AMPK on endogenous glucose production are factious (26). Many studies have concluded that AMPK action inhibits rather than promotes glucose release from the liver (26). As such, the current study employed in vivo 2 H/ 13 C metabolic flux analysis (MFA) to characterize the dynamic alterations in nutrient fluxes contributing to endogenous glucose production in C57Bl/6J mice in response to an acute treadmill running bout. Subsequently, liver-specific AMPK␣1␣2 KO and WT littermates were used to test the hypothesis that AMPK is required for the increment in hepatic glucose producing and citric acid-cycle (CAC)-related fluxes during exercise. Mice lacking hepatic AMPK catalytic subunits exhibited an impaired ability to maintain glucose homeostasis during a 30-min treadmill run. This was associated with lower liver glycogen and glycogenolytic response to exercise. Gluconeogenesis-related fluxes during sedentary and exercise conditions were not altered by deletion of hepatic AMPK␣1␣2.

Exercise promotes changes in circulating hormones and liver energy metabolism
Initial studies characterized the exercise response in C57Bl/6J mice during an acute treadmill running bout at 45% of maximal running speed. Body weights of sedentary (SED) and treadmillrunning (exercising; EX) mice were comparable (Table 1). Maximal running speed was similar between the SED and EX groups at 36.5 Ϯ 1.3 and 38.5 Ϯ 1.2 m min Ϫ1 , respectively (Table 1). During stable isotope infusions, SED mice remained on a stationary treadmill, whereas the exercising mice ran on a treadmill at 17.3 Ϯ 1.3 m min Ϫ1 for 30 min (Table 1). Characteristic hormonal responses to exercise were observed in a separate cohort of mice undergoing the study conditions. Arterial insulin declined following the onset of exercise (Table 1). Plasma glucagon was elevated at 30 min of treadmill running (Table 1). Liver glycogen was reduced in the EX group at the conclusion of the treadmill run ( Table 1). The exercise bout promoted changes in hepatic energy status. Liver ATP was lower and AMP was higher in EX mice ( Table 1). The AMP/ATP ratio was increased and energy charge was decreased in EX mice (Table  1). It is important to note that liver ATP relative to ADP and AMP are lower in SED mice than that reported in perfused livers (27,28). Livers in this study were harvested and freezeclamped in under 20 s following euthanization. It is possible that compromised oxygen availability during that time frame may have induced energy discharge. 2 H/ 13 C MFA allowed glucose-producing and associated nutrient fluxes to be determined in the SED and EX mice. The studies achieved isotopic steady state in sedentary mice and a new isotopic dynamic steady state during exercise conditions. The time course of the mass isotopomer distribution (MID) for the fragment ion range m/z 301-305 is provided in Fig. 1. The remaining five fragment ions of glucose derivatives used for 2 H/ 13 C MFA also show isotopic steady state (data not shown). Pronounced disparities between the SED and EX mice in the glucose-producing fluxes estimated in vivo ( Fig. 2A) were identified. Blood glucose was higher at 10 min and lower at 30 min in EX mice (Fig. 2B). Treadmill running enhanced endogenous glucose production (V EndoRa ) at the 10-, 20-, and 30-min time points (Fig. 2C). Fluxes from glycogen (V PYGL ; Fig. 2D) and gluconeogenic sources (V Aldo ; Fig. 2E) were elevated with exercise. Interestingly, the increased gluconeogenic flux was due to substrate provision from phosphoenolpyruvate (PEP; V Enol ) but not glycerol (V GK ) (Fig. 2, F and G).

Treadmill running stimulates hepatic nutrient fluxes in vivo
In addition to glucose synthesis, CAC-related reactions were estimated at rest and during treadmill running. The return of PEP to pyruvate increased during exercise (V PKϩME ; Fig. 2H). Flux from unlabeled substrates to pyruvate (V LDH ) was ϳ1.7fold higher throughout the running bout (Fig. 2I), which contributed to the increase in anaplerotic flux from pyruvate to oxaloacetate (V PC ; Fig. 2J). Cataplerosis (V PCK ) exhibited an increment at each time point during exercise (Fig. 2K). Similarly, flux from oxaloacetate and acetyl-CoA to citrate (V CS ), flux from propionyl-CoA to succinyl-CoA (V PCC ), and flux from succinyl-CoA to oxaloacetate (V SDH ) were greater in the EX group compared with the SED group (Fig. 2, L-N). Of note, V PCC was higher at the 30-min time point compared with the 0-min time point in the SED mice (Fig. 2M).
Assessment of the relative contributions of glucose sources to V EndoRa showed that the increase in V PYGL throughout the 30-min Table 1 Biometric characteristics of sedentary and treadmill-running C57Bl/6J mice Shown are body weight (g) before food and water withdrawal at Ϫ510 min in SED and EX mice as well as maximal running speed (m min Ϫ1 ) and experimental running speed (45% of maximal running speed for EX group; m min Ϫ1 ). Liver glycogen (mg g Ϫ1 ), liver adenine nucleotides (TAN ϭ ATP ϩ ADP ϩ AMP; mol g Ϫ1 ), and AMP/ATP and energy charge ([ATP ϩ 0.5 ADP]/[TAN]) were determined in SED and EX mice. Plasma insulin and glucagon at 10 min before commencing the acute exercise bout and 30 min of treadmill running at 45% of maximal running speed were obtained from a separate cohort of age-matched, male C57Bl/6J mice. Data are mean Ϯ S.E. n ϭ 5-9 mice/group.

Role of hepatic AMPK in glucose homeostasis during exercise
treadmill run was associated with a higher V PYGL /V EndoRa ratio (Fig. 2O). The greater contribution of glycogen to V EndoRa mediated a decline in the contribution of gluconeogenesis to V EndoRa , as indicated by the V Aldo /V EndoRa ratio at the 10-, 20-, and 30-min time points in the EX mice (Fig. 2P). It should also be noted that plasma non-esterified fatty acids (NEFA) were similar between groups (Fig. 2Q).

Loss of AMPK does not influence running capacity
Previous work has shown that treadmill running increases liver AMPK phosphorylation (11). Initial work aimed to assess the role of hepatic AMPK␣ on functional parameters. WT mice and littermates with a liver-specific deletion of AMPK ␣1 and

AMPK is required for glucose homeostasis during exercise
Whereas mice of both genotypes performed exercise at the same intensity, the absence of hepatic AMPK caused marked alterations in estimated glucose-related fluxes (Fig. 3A). Plasma NEFAs increased similarly while running at 45% of maximal running speed in WT and KO mice (Fig. 3B). Under sedentary conditions, blood glucose levels were comparable between groups; however, KO mice exhibited a reduced arterial glucose during exercise (Fig. 3C). The increment in V EndoRa was blunted in the KO mice (Fig. 3D). The lower endogenous glucose production during exercise in the KO group was accompanied by reduced V PYGL (Fig. 3E). Fluxes of reactions involved in gluconeogenesis, V Aldo , V GK , and V Enol , were similar between groups at rest and during the treadmill run (Fig. 3, F-H). CAC-related fluxes were comparable between sedentary and exercising WT and KO mice. Pyruvate cycling (V PKϩME ) was elevated during the treadmill running (Fig. 3I) in both groups. Anaplerosis and related fluxes (V LDH , V PC , and V PCC ) were raised with exercise in WT and KO mice (Fig. 3, J, K, and N). The generation of PEP (V PCK ) was similar between genotypes before exercise (Fig. 3L). During treadmill running, the increase in V PCK was comparable in both groups (Fig. 3L). CAC fluxes, V CS and V SDH , showed similar increases throughout the treadmill running in WT and KO mice (Fig. 3, M and O).

AMPK influences liver nutrient partitioning and energy status under sedentary, short-term fasted conditions
Changes in hepatic molecular mediators of metabolism and metabolites were observed in the absence of differences in liver nutrient fluxes between genotypes following 8.5 h of food and water withdrawal. Immunoblotting was performed to confirm deletion of AMPK␣ subunits (Fig. 4A). Glycogen synthase phosphorylation was similar in both groups (Fig. 4B); however, UDP-glucose pyrophosphorylase 2 (UGP2) was reduced (Fig.  4C). This was associated with a decline in liver glycogen levels in KO mice (Fig. 4D). Mice with a deletion of AMPK ␣1 and ␣2 subunits displayed a reduction in liver ATP (

Altered metabolites are retained following exercise in mice lacking liver AMPK
Deletion of hepatic AMPK␣ subunits was confirmed by immunoblotting (Fig. 5A). Liver glycogen was reduced in the KO mice following 30 min of exercise (Fig. 5B). Energy charge were similar between groups (Fig. 5C). Liver ATP was comparable between genotypes (

AMPK affects liver nutrient partitioning and energy status in long-term fasted and refed mice
The lower glycogen before the exercise bout in KO mice prompted fasting-refeeding experiments to provide insight into the ability of mice lacking liver AMPK␣1␣2 to synthesize liver glycogen. Deletion of hepatic AMPK␣ subunits is represented by immunoblotting (Fig. 6A). Liver glycogen was depleted and comparable between genotypes following 24 h of food withdrawal (Fig. 6B). Liver glycogen increased in both groups following 6 h of refeeding compared with fasted levels (Fig. 6B). However, the increase in glycogen stores were attenuated in the KO mice compared with the WT mice in the refed condition (Fig. 6B). Energy charge was comparable between groups (

Role of hepatic AMPK in glucose homeostasis during exercise
with the 24 h of fasting (Fig. 6G). Triglycerides were lower in KO mice following the 24-h fast compared with WT mice (Fig.  6H). Within the WT mice, the refed conditions exhibited lower liver triglycerides (Fig. 6H), diglycerides (Fig. 6I), phospholipids (Fig. 6J), and total cholesterol esters (Fig. 6K). Cholesterol esters were higher in livers lacking AMPK␣ compared with control livers in the 24-h fasted conditions (Fig. 6K). 6 h of food availability following the long-term fast lowered cholesterol esters in the KO mice (Fig. 6K).

Mice lacking hepatic AMPK␣1␣2 subunits showed changes in fibroblast growth factor 21 (FGF21) physiology
Previous studies in short-term fasted mice lacking liver AMPK␣ showed metabolic characteristics of long-term food withdrawal (29). FGF21 is secreted by the liver in response to starvation and exercise (30,31). Furthermore, genetically engineered mice indicate that FGF21 promotes gluconeogenesis (32). Interestingly, livers obtained from sedentary mice in the treadmill at the 0-min time point displayed lower hepatic FGF21 protein as determined by immunoblotting in KO mice compared with the WT mice ( Fig. 7A). Hepatic FGF21 was reduced in KO mice compared with WT mice following treadmill running as well (Fig. 7B). To exclude the possibility that the treadmill belt and enclosure may cause a unique FGF21 response, a separate cohort of sedentary, 9-h fasted mice were studied outside the treadmill in a 1-liter plastic container with bedding. This cohort of KO mice displayed elevated liver FGF21 mRNA (Fig. 7C), reduced liver FGF21 protein (Fig. 7D), and increased plasma FGF21 (Fig. 7E) in relation to WT mice. Deletion of hepatic AMPK␣ subunits was confirmed by immunoblotting (Fig. 7F). The increased plasma FGF21 in KO mice was paralleled by a paradoxical decline in potential mediators of FGF21 action. Hepatic phosphorylation of MEK1/2 ( Fig. 7G) and its downstream target, ERK1/2 ( Fig. 7H), was reduced in KO mice. Phosphorylation of ERK substrate, the p90 ribosomal S6 kinases (p90RSK), was not statistically different between groups (

Discussion
The predominant hypothesis concerning AMPK's role in the control of endogenous glucose production is that it impedes glucose release from the liver. Deletion of the primary AMPK kinase, liver kinase B1, in the liver of mice results in hyperglycemia (33). Moreover, LKB/AMPK-associated signaling pathways have been reported to negatively regulate the expression of genes involved in glucose production, such as glucose-6phosphatase (G6Pase) and phosphoenolpyruvate carboxykinase (PEPCK) (33)(34)(35)(36). Support for the aforementioned hypothesis also arises from the experimental use of biguanides and 5-aminoimidazole-4-carboxamide 1-␤-D-ribofuranoside (AICAR). These pharmacological agents concur-rently activate AMPK and lower glucose release from hepatocytes or blood glucose in vivo (22,37,38). Although more recently a growing body of research indicates that the ability of biguanides and AICAR to inhibit hepatic glucose production occurs in an AMPK-independent manner (22,39,40).
Similar to these pharmaceutical agents, exercise promotes energy discharge and an increase in AMPK phosphorylation (11). However, this physiological stressor is associated with an increase in glucose release from the liver. The current study aimed to test the requirement for hepatic AMPK to maintain glucose homeostasis through the promotion of endogenous glucose production and associated nutrient fluxes during exercise. A recently developed in vivo 2 H/ 13 C MFA technique (41) was used to characterize intermediary metabolism fluxes in the conscious C57Bl/6J mouse during short-term treadmill running at 45% of their maximal running speed. Further experiments utilized mice with a liver-specific deletion of AMPK␣1␣2 subunits. The results indicate that deletion of liver AMPK ␣ subunits in mice impairs glycogen deposition. This compromises the ability of these mice to increase glycogenolysis (V PYGL ) as needed to maintain euglycemia. Loss of hepatic AMPK catalytic subunits results in a decline in ATP and increase in AMP/ATP ratio; however, gluconeogenic and CACrelated responses to the 30-min treadmill running bout are not influenced by genotype.
Under sedentary conditions, arterial glucose concentration and endogenous glucose production were comparable between WT and KO mice. This is consistent with previous studies in which AMPK␣ and AMPK␤ subunits were deleted (22,39,42,43). During an acute exercise bout, endogenous glucose production (V EndoRa ) and blood glucose were lower in KO mice. The diminished increment in V PYGL in response to treadmill running in mice lacking hepatic AMPK␣1␣2 indicates that lower glycogenolysis contributed to reduced endogenous glucose production and impaired glucose homeostasis during muscular work.
In vitro work has shown that AMPK phosphorylates protein targeting to glycogen (R5/PTG), promoting the degradation of this PP1-interacting protein in a laforin-malin-dependent manner (44). Also, AMPK catalytic activity positively regulates the formation of the laforin-malin complex (45). These AMPK actions could facilitate greater glycogen mobilization and contribute to the enhanced glycogenolysis during exercise in mice with hepatic AMPK. However, the reduction in glycogen flux to glucose 6-phosphate precipitated during exercise in KO mice was associated with diminished hepatic glycogen before commencing exercise (8.5-h fast) and upon completion of the 30-min treadmill run. As such, the lower glycogenolytic response to exercise is more likely the result of limited glycogen as a substrate.

Role of hepatic AMPK in glucose homeostasis during exercise
Glycogen content is a product of formation and degradation processes. The metabolic cycle of UDP-glucose insertion into glycogen granules and generation of glucose 6-phosphate from glycogen is centered around glycogen synthase and glycogen phosphorylase function, respectively (46). Hepatic glycogen was comparable between genotypes following a 24-h fast. However, a reduction in hepatic glycogen content was observed in AMPK␣1␣2-deficient livers from 24-h fasted, 6-h refed mice. This suggests that altered hepatic glycogen deposition occurs in the absence of AMPK␣1␣2. Phosphorylation of the (Ser-641) phosphorylation site, which inhibits enzymatic activity, was not different between genotypes. It is conceivable that another phosphorylation site with inhibitory action is increased (47).
In addition to covalent modification of glycogen synthase, alterations in energy status resulting from loss of AMPK␣1␣2 hold the potential to influence glycogen synthase activity via allosteric regulation. Glycogen synthase activity diminishes with declines in adenylate energy charge (48). KO mice had lower hepatic energy charge before the onset of exercise. Previous work showed that the energy discharge exhibited by hepatic AMPK deletion was associated with lower glycogen accumulation in response to AICAR infusion (22). In contrast, fastedrefed mice showed that absolute concentrations of hepatic adenine nucleotides and relative energy state ratios were comparable between genotypes despite the persistent reduction in liver glycogen content in AMPK-null livers. These results suggest that the lower glycogen is not mediated by changes in energy state.
Previous studies have identified glucokinase protein as being reduced in hepatocytes lacking AMPK␣1␣2 (49). This may influence substrate for glycogen deposition. In the current study, a reduction in the provision of glycogen precursors could impact glycogen storage. Liver UGP2 protein was reduced in KO mice. UGP is an enzyme that catalyzes the generation of UDP-glucose, the immediate precursor for glycogen formation (50). Mice overexpressing and lacking AMPK␥3 exhibit an increase and decrease in UGP transcript levels, respectively (51). Muscle glycogen is positively correlated to UGP and AMPK in swine with mutated activation of AMPK␥ subunits (52)(53)(54). The roles of covalent modification, allosteric control, and substrate availability in mediating the alterations in liver glycogen homeostasis in the absence of AMPK remain to be fully clarified.
As indicated previously, many in vivo and in vitro studies employing genetic tools have implicated AMPK as a negative regulator of gluconeogenesis. This hypothesis is supported by findings in mice with short-term overexpression of constitutively active AMPK␣2 in the liver (34). PEPCK and G6Pase mRNA were reduced 48 h following adenovirus-mediated gene transfer of constitutively active AMPK␣2 in the liver of mice (34). These molecular changes were associated with lower blood glucose (34). In agreement, 12-week-old mice with hepatic deletion of AMPK␣2 displayed elevated endogenous glucose production following 6 h of fasting (55). Interestingly, PEPCK and G6Pase mRNA were not altered by loss of AMPK␣2; however, the enzyme activities were elevated (55). In contrast, recent work has added to the growing body of literature refuting such a regulatory role for liver AMPK in glucose homeostasis. We recently reported that 15-week-old mice lacking liver AMPK␣1␣2 subunits had no effect on endogenous glucose production or gluconeogenic flux following 7 h of fasting (29).
The present study further evaluated the influence of AMPK on gluconeogenesis under acute physiological conditions (8.5-9 h of fasting and acute exercise) in 15-16-week-old mice. WT and KO mice exhibited comparable contributions of PEP and glycerol to glucose production, suggesting that AMPK does not exert control over gluconeogenesis. Means of gluconeogenic regulation include the efficiency of intrahepatic conversion of gluconeogenic precursors to glucose as well as substrate supply and hepatic gluconeogenic precursor extraction (9). The absence of an increase in gluconeogenesis in sedentary, postabsorptive mice lacking hepatic AMPK␣1␣2 may be related to the altered energy state attenuating the efficacy of gluconeogenic reactions. Our previous work identified integrative mitochondrial oxidative phosphorylation as being impaired in livers lacking AMPK (22). The current study showed a reduction in ATP, TAN, and energy charge in KO mice. The impaired capacity for ATP provision may have prevented uncontrolled CAC-related and gluconeogenic reactions during sedentary conditions. During exercise, hepatic extraction of gluconeogenic precursors may create an abundance of gluconeogenic precursor availability, which may have masked the molecular control AMPK has over CAC-related and gluconeogenic fluxes.
Our previous work using [ 13 C 3 ]propionate at the same dose as in the present study shows no effect on cataplerosis (V PCK ) or gluconeogenic flux from PEP (V Enol ) (41). Propionate is a precursor of succinyl-CoA. The 2 H/ 13 C MFA is sufficiently robust to detect an increase in CAC fluxes (V CS and V SDH ) (41). It is important to note that using this methodology, a higher V CS in liver-specific AMPK KO mice compared with WT mice following a 19-h fast was shown (29). Thus, the method is sensitive to experimental and physiological differences in CAC and gluconeogenic fluxes between WT and KO mice. We could use a lower-dose [ 13 C 3 ]propionate if it were necessary, but there would be lower confidence in the flux estimates (41).
KO livers were shown to have reduced glycogen and increased triglycerides at 8.5 h but not 24 h of fasting. The short-term fasting phenotype resembles that of a longer fast (29), suggesting that the KO mice transition to the long-term fasted state quicker than the WT mice following food withdrawal. Extended fasting promotes the expression and secretion of FGF21 from the liver (56). Moreover, FGF21 stimulates gluconeogenesis in the fasted state when insulin is low (32,57).

Role of hepatic AMPK in glucose homeostasis during exercise
We hypothesized that hepatic FGF21 action may compensate for the absence of AMPK and preserve appropriate gluconeogenic response in KO mice. Indeed, liver FGF21 mRNA and plasma FGF21 were higher in KO mice. Interestingly, however, FGF21 action seems to be attenuated in KO livers, as indicated by a reduction in MEK1/2 and ERK1/2 phosphorylation. Of note, constitutive activation of hepatic MEK1 and deletion of hepatic ERK1/2 increases and decreases blood glucose in mice,

Role of hepatic AMPK in glucose homeostasis during exercise
respectively (58). In light of these results, loss of AMPK may impede hepatic MEK/ERK phosphorylation by FGF21 and, subsequently, prevent increased gluconeogenesis. FGF21 administration has been shown to induce AMPK phosphorylation and regulate similar metabolic pathways (59, 60). Alternatively, AMPK and FGF21 action may converge on MEK/ERK signaling. Previous studies have reported that under conditions of energetic stress, such as that induced by AICAR, AMPK is upstream and required for full phosphorylation and activation of the MEK/ ERK signaling pathway (61,62). The apparent FGF21 resistance in livers with attenuated AMPK activation bears a striking resemblance to phenotypes observed in obesity and diabetes models (13,63).
The use of stable isotopes to probe metabolic systems has strengths and limitations (64,65). We have performed extensive characterization of 2 H/ 13 C MFA using different doses of [ 13 C 3 ]propionate (41). We showed that a range of [ 13 C 3 ]propionate doses have no effects on plasma glucose and hepatic glucose production. This is in agreement with previous studies (66). With this foundation, we used [ 13 C 3 ]propionate in the present study to assess the requirement of AMPK for glucose production and associated nutrient fluxes during exercise. Indeed, propionate can increase pyruvate carboxylase activity (67). Perry et al. (68) recently reported that combining intra-arterial infusion of propionate with an intra-arterial infusion of mass quantities of a precursor for this enzyme ([3-13 C]lactate) increases CAC interme-

Role of hepatic AMPK in glucose homeostasis during exercise
diates, endogenous glucose production, and circulating glucose, in the presence of a 6.5-fold increase in plasma insulin. Such effects do not occur with propionate infusion alone. This is consistent with the absence of reported plasma glucose or insulin when they gave propionate alone (68). Clearly combining intra-arterial propionate with an exogenous intra-arterial lactate load causes profound insulin resistance when the metabolic demand of the organism does not match the increment in anaplerotic flux. A confounding aspect of the experimental design of Perry et al. (68) is that propionate and lactate were infused into a carotid artery catheter, where it has a natural first pass to the contralateral carotid artery and other arteries that perfuse the brain. It is possible that the solutes or volume load of multiple infusates can evoke a response that contributes to the effects of combining propionate with a lactate load.
Energy status is an important regulator of glucose metabolism (69,70). This has prompted substantial scientific investigation into the energy sensor, AMPK, to understand its role in metabolic control (71). The studies presented here aimed to evaluate the AMPK-dependent actions on endogenous glucose production during the physiological energy stress of acute exer-cise. Initial studies employed 2 H/ 13 C MFA in C57Bl/6J mice to determine glucose-producing and -associated nutrient fluxes throughout a 30-min treadmill run. Endogenous glucose production (V EndoRa ), glycogenolysis (V PYGL ), and gluconeogenesis from PEP (V Enol ) were increased at 10, 20, and 30 min of exercise. Concurrent increases in cataplerosis (V PCK ), CAC fluxes (V CS and V SDH ), and anaplerosis-related fluxes (V PC , V PCC , and V LDH ) were shown. Subsequent exercise experiments used mice with a liver-specific deletion of AMPK␣ subunits. AMPK KO mice were unable to maintain euglycemia during treadmill running. V EndoRa and V PYGL were lower in KO mice. However, V Enol and downstream fluxes were comparable between WT and KO mice throughout muscular work. In conclusion, hepatic AMPK is necessary for the glycogenolytic, but not the gluconeogenic, response to exercise.

Animal models
Procedures were approved by the Vanderbilt University Animal Care and Use Committee. 13-Week-old male C57Bl/6J

Role of hepatic AMPK in glucose homeostasis during exercise
mice were purchased from the Jackson Laboratory (Bar Harbor, ME) and underwent surgical procedures at 14 weeks of age. Mice exhibiting a liver-specific knock-out of AMPK ␣1 and ␣2 subunits were generated by crossing mice containing floxed AMPK ␣1 and ␣2 subunits on a C57Bl/6J background with AlbCre mice (B6.Cg-Tg(Alb-cre)21Mgn/J). Liver-specific dele-tion of AMPK ␣1 and ␣2 subunits was assessed by genotyping and/or immunoblotting. Floxed AMPK␣1␣2 mice were used as WT mice. Mice underwent surgical procedures at 14 weeks of age if used for stable isotope infusion studies or for experiments requiring the acquisition of arterial samples. All mice received PicoLab Laboratory Rodent Diet (5L0D, Purina, Richmond, IN) and had free access to water in temperature-and humiditycontrolled housing maintained on a 12-h/12-h light/dark cycle.

Exercise stress test
An exercise stress test was performed to determine maximal running speed, as described previously (72). Briefly, mice were acclimatized to running on an enclosed single-lane treadmill by completing a 10-min exercise bout at 10 m min Ϫ1 (0% incline) a day before the exercise stress test and 2 days before surgical procedures. To determine maximal running speed, mice were placed in an enclosed single-lane treadmill (Columbus Instruments, Columbus, OH). Following a 10-min sedentary period, mice initiated running at 10 m min Ϫ1 (0% incline). The treadmill speed was increased every 3 min by 4 m min Ϫ1 until mice reached exhaustion. Exhaustion was defined as the time point when mice remained on the shock grid at the back of the treadmill for Ͼ5 s.

Surgical procedures
At 14 weeks of age, mice underwent catheterization procedures, as performed previously (73). Briefly, the jugular vein and carotid artery were catheterized for venous infusions and arterial sampling, respectively. The free end of each catheter was exteriorized at the back of the neck, flushed with 200 units/ml heparinized saline and 5 mg/ml ampicillin, and sealed with stainless steel plugs. Following surgery, the mice were housed individually and allowed 7-8 days of postoperative recovery before an acclimation treadmill running bout of 10 min at 45% of their predetermined maximal running speed (0% incline). Two days later, the stable isotope infusion studies were performed. All mice were within 10% of presurgical weight on the day of stable isotope infusions. An age-matched cohort of WT and KO mice had only arterial catheters inserted for blood glucose, plasma insulin, and plasma FGF21 measurements at 9 h of fasting. These mice did not undergo treadmill experiments.

Stable isotope infusions
See Fig. 8 for a schematic representation of the in vivo infusion procedures and timeline. Within an hour of the start of the light cycle, food and water were withdrawn for the remainder of the study (9 h). Four hours into the fast, mice were placed in an enclosed single-lane treadmill (Columbus Instruments), and the exteriorized catheters were connected to catheter leads and infusion syringes. At 5 h into the fast, an 80-l arterial blood sample was obtained to determine natural isotopic enrichment of plasma glucose. Immediately following this sample, a quan-titative stable isotope delivery to increase isotopic enrichment above natural isotopic labeling was initiated as performed previously (41 Four 100-l arterial blood samples were obtained 90 -120 min following the [ 13 C 3 ]propionate bolus (time ϭ 0 -30 min) to determine arterial blood glucose and plasma NEFA levels as well as to perform 2 H/ 13 C MFA to determine endogenous glucose and associated nutrient fluxes. The sample taken at 90 min following the [ 13 C 3 ]propionate bolus (time ϭ 0 min) was obtained while mice were in a sedentary state on a stationary treadmill. Samples taken 100 -120 min following the [ 13 C 3 ]propionate bolus (time ϭ 10 -30 min) were obtained while mice were in a sedentary state or performing an acute treadmill running bout at 45% of their predetermined maximal running speed. Plasma samples were stored at Ϫ20°C until analysis. Donor erythrocytes were provided throughout the duration of the study to ensure that hematocrit did not fall Ͼ10%. Immediately following the final sample, mice were removed from the treadmill and sacrificed by cervical dislocation. Liver tissues were rapidly excised, freeze-clamped in liquid nitrogen, and stored at Ϫ80°C until analysis. Stable isotopes were purchased from Cambridge Isotope Laboratories, Inc. (Tewksbury, MA). Each infusate was prepared in a 4.5% 2 H 2 O-enriched saline solution unless otherwise specified.

Glucose derivatization and GC-MS analysis
40 l of plasma from the Ϫ210-, 0-, 10-, 20-, and 30-min time points was processed to obtain di-O-isopropylidene propionate, aldonitrile pentapropionate, and methyloxime pentapropionate derivatives of glucose as described previously (41,74). GC-MS analysis was completed as described previously (41) with minor modifications. A custom MATLAB function was used to integrate derivative peaks to obtain MIDs for six fragment ions. The following fragment ion ranges were used for determining  A, liver FGF21 as determined by immunoblotting and representative immunoblot of hepatic FGF21 and ␤-actin. Livers taken at time ϭ 0 min from non-catheterized, WT mice and mice exhibiting a liver-specific deletion of AMPK ␣1 and ␣2 subunits (KO) placed on an enclosed, stationary treadmill. B, liver FGF21 as determined by immunoblotting and representative immunoblot of hepatic FGF21 and ␤-actin. Livers were taken following stable isotope infusions and 30 min of treadmill running at 45% of maximal running speed from catheterized WT and KO mice. A separate cohort of sedentary, catheterized WT and KO mice fasted for 9 h, not in an enclosed treadmill, was studied to evaluate FGF21 characteristics. C, liver FGF21 mRNA. D, liver FGF21 as determined by immunoblotting and representative immunoblot of hepatic FGF21 and ␤-actin. E, plasma FGF21. F, liver AMPK␣ as determined by immunoblotting and representative immunoblot of hepatic AMPK␣ and ␤-actin. G, liver phosphorylated MEK1/2-to-total MEK1/2 ratio (pMEK S217/221 /MEK1/2) as determined by immunoblotting in and representative immunoblot of hepatic pMEK S217/221 , total MEK1/2, and ␤-actin. H, liver phosphorylated ERK1/2-to-total ERK1/2 ratio (pERK T202/Y204 /ERK1/2) as determined by immunoblotting in and representative immunoblot of hepatic pERK T202/Y204 , total ERK1/2, and ␤-actin. I, liver phosphorylated p90RSK/total p90RSK ratio (pp90RSK S380 /p90RSK) as determined by immunoblotting and representative immunoblot of hepatic pp90RSK S380 , p90RSK, and ␤-actin. J, plasma glucose (mM). K, plasma insulin (ng ml Ϫ1 ). Data are mean Ϯ S.E. (error bars); n ϭ 6 -11 mice/group. *, p Ͻ 0.05 versus WT.

Role of hepatic AMPK in glucose homeostasis during exercise 2 H/ 13 C metabolic flux analysis
A detailed description of the in vivo metabolic flux analysis methodology employed in these studies has been provided (41). Briefly, a reaction network was constructed using the INCA software package (75) (http://mfa.vueinnovations.com/mfa). 3 The reaction network defined the carbon and hydrogen transitions for biochemical reactions linking hepatic glucose production and associated intermediary metabolism reactions. Given the model functions based on steady-state isotopomer balances, it does not depend on metabolite pool sizes. Flux through each network reaction was estimated relative to V CS (fixed at 100) by minimizing the sum of squared residuals between simulated and experimentally determined MIDs of the six fragment ions described previously (41,76). The formulation of the isotopomer balance equations used to simulate the fragment MIDs has been thoroughly detailed by Antoniewicz et al. (77). Flux estimates were repeated 50 times from random initial values. Goodness of fit was assessed by a 2 test. Confidence intervals of 95% were determined as described previously (41,76). Fits were accepted according to a 2 test (p ϭ 0.05) with 34 degrees of freedom. The infusion rate of [6,6-2 H 2 ]glucose (V Inf ) and mouse weights were used to determine absolute values.

Blood, plasma, and tissue metabolite analyses
Arterial blood glucose was measured using an Accu-Chek® glucometer (Roche Diagnostics). Plasma NEFAs were measured via an enzyme-linked spectrophotometric assay (NEFA C kit, Wako Chemicals USA Inc., Richmond, VA). Liver glycogen was determined enzymatically from tissue homogenates as described previously (78). Hepatic adenine nucleotides were extracted and measured as performed previously (22). ). Liver lipids, including triglycerides, diglycerides, phospholipids, and cholesterol esters, were measured as described previously (29). A cohort of 15-16-week-old, uncatheterized, male WT and KO mice were placed in an enclosed, stationary single-lane treadmill (Columbus Instruments) 4 h into a fast and were sacrificed for liver glycogen, lipids, adenine nucleotides, and immunoblotting in the sedentary state at 8.5 h of fasting (equivalent to the time ϭ 0 min of infusion studies). A cohort of 15-16-week-old, uncatheterized, male WT and KO mice were 24-h fasted and 24-h fasted-6-h refed. These mice were sacrificed to evaluate liver glycogen, lipids, and adenine nucleotides. A separate cohort of age-matched, male C57Bl/6J mice were studied to determine plasma insulin and glucagon concentrations in response to exercise. Plasma insulin was evaluated by a double antibody method as performed previously (79). Plasma glucagon was assessed as reported (80). Stable isotope infusions and an acute treadmill running bout were performed following 9 -10 days of recovery from arterial and jugular catheterization surgeries. At 210 min before the treadmill running bout, a 2 H 2 O bolus was administered into the venous circulation to enrich total body water at 4.5%. Simultaneously, a 440 mol kg Ϫ1 min Ϫ1 [6,6-2 H 2 ]glucose prime was infused followed by a continuous 4.4 mol kg Ϫ1 min Ϫ1 [6,6-2 H 2 ]glucose infusion. Ninety minutes before the onset of exercise, a 1.1 mmol kg Ϫ1 min Ϫ1 prime and 0.055 mmol kg Ϫ1 min Ϫ1 continuous infusion of [ 13 C 3 ]propionate was initiated. Donor erythrocytes were administered to prevent a decline in hematocrit. Arterial samples were obtained before stable isotope infusion as well as during 30-min sedentary and treadmill running (45% of maximal running speed) conditions for 2 H/ 13 C metabolic flux analysis of hepatic intermediary metabolism.

Role of hepatic AMPK in glucose homeostasis during exercise
Immunoblotting Liver tissue was homogenized in an extraction buffer containing 20 mM NaCl, 20 mM Tris-HCl, 0.1 mM EDTA, 1% Triton X-100, 0.5% sodium deoxycholate, 0.1% ␤-mercaptoethanol (pH 7.4), and a protease and phosphatase inhibitor mixture (Thermo Fisher Scientific). Liver homogenate was centrifuged (15 min at 800 ϫ g and 4°C), and the supernatant protein concentration was assessed using the Bradford method (Bio-Rad). Liver (20 -40 g) proteins were denatured and reduced at 70°C, separated on a NuPAGE 4 -12% BisTris gel (Invitrogen), and transferred to a PVDF membrane. Membranes were probed with the following antibodies: phospho-acetyl-CoA carboxylase ( Following incubation with an HRP-conjugated secondary antibody, the membranes were exposed to a chemiluminescent HRP-substrate (Millipore, Billerica, MA) and imaged using a ChemiDoc TM imaging system and Image Lab TM software (Bio-Rad). Densitometry was completed using ImageJ software.

Real-time PCR
Liver RNA extraction and reverse transcription were performed as described (81). Real-time PCR was completed with FGF21 and housekeeping gene, ␤-actin, TaqMan gene expression assays (Thermo Fisher Scientific). Data were analyzed using the 2 Ϫ⌬⌬Ct method (82).

Statistical analyses
SigmaStat statistical software was used to perform twotailed t tests, ANOVA, one-way repeated measures ANOVA, and two-way repeated measures ANOVA as appropriate to detect statistical differences (p Ͻ 0.05) followed by Bonferroni post hoc tests. Comparisons of interest were SED versus EX and WT versus KO. All data are reported as means Ϯ S.E. analyzed and interpreted data. C. C. H. drafted the manuscript. All authors contributed to revising the manuscript for critically important intellectual content. All authors approved the manuscript for publication.