AMP‐activated protein kinase activators have compound and concentration‐specific effects on brain metabolism

AMP‐activated protein kinase (AMPK) is a key sensor of energy balance playing important roles in the balancing of anabolic and catabolic activities. The high energy demands of the brain and its limited capacity to store energy indicate that AMPK may play a significant role in brain metabolism. Here, we activated AMPK in guinea pig cortical tissue slices, both directly with A769662 and PF 06409577 and indirectly with AICAR and metformin. We studied the resultant metabolism of [1‐13C]glucose and [1,2‐13C]acetate using NMR spectroscopy. We found distinct activator concentration‐dependent effects on metabolism, which ranged from decreased metabolic pool sizes at EC50 activator concentrations with no expected stimulation in glycolytic flux to increased aerobic glycolysis and decreased pyruvate metabolism with certain activators. Further, activation with direct versus indirect activators produced distinct metabolic outcomes at both low (EC50) and higher (EC50 × 10) concentrations. Specific direct activation of β1‐containing AMPK isoforms with PF 06409577 resulted in increased Krebs cycle activity, restoring pyruvate metabolism while A769662 increased lactate and alanine production, as well as labelling of citrate and glutamine. These results reveal a complex metabolic response to AMPK activators in brain beyond increased aerobic glycolysis and indicate that further research is warranted into their concentration‐ and mechanism‐dependent impact.

levels of α1 expression, but this subunit was not detected above background levels elsewhere (Turnley et al., 1999).The α1 subunit is not expressed in astrocytes, but the α2, β2 and to a lesser extent β1 are expressed in these cells, with α2 expression being predominantly cytoplasmic.Cortical neurons, by contrast, showed differential expression of β1 or β2 with some expressing both forms.Less is known about the roles of the individual γ subunits in brain, although γ1 and to a lesser extent γ2 expression has been reported in neurons (Turnley et al., 1999).The affinity of the nucleotide-binding sites in the different γ subunits has been suggested to drive the difference in the allosteric response to AMP by each isoform, with greater response reported by γ1 > γ2 > γ3 (Carling, 2017).A series of possible post-translational modifications also adds to the complexity of possible responses that AMPK can deliver (Ovens et al., 2021), as well as different intracellular localisation (lysosomal, Zhang et al., 2022;mitochondrial, Drake et al., 2021) and direct complexing with enzymes (Lee et al., 2015).
The γ subunit of AMPK possesses three functional binding sites for adenosine nucleotides (Xiao et al., 2007).While the third of these is generally permanently occupied by AMP, the other two may bind any of ATP, ADP or AMP, depending on the local concentrations of each.When energy availability is low, ADP and/or AMP allosterically and progressively displace ATP on the γ subunit, which exposes the catalytic domain in the α-subunit.This allows upstream kinases Ca 2+ /calmodulin-activated protein kinase kinaseβ (CaMKK2;E.C. 2.7.11.17) and liver kinase B1 (LKB1; E.C. 2.7.11.1) to phosphorylate the β subunit at Thr172 to activate AMPK (Ronnett et al., 2009;Woods et al., 2003;Xie et al., 2006).The degree of displacement of ATP binding on the γ subunit controls the degree of activation of AMPK, with 100-fold activation following the displacement of one ATP and a further 10-fold activation on the displacement of the second ATP.When bound, both AMP and ADP inhibit dephosphorylation of Thr 172 (Davies et al., 1995;Xiao et al., 2011).This sensitivity and responsiveness to cellular AMP/ADP/ATP balance positions AMPK as a key regulator of cellular metabolic activity (Hardie et al., 2012).This may be further enhanced by binding proximate to the carbohydrate-binding module (also termed glycogen binding domain) of AMPK (Xiao et al., 2013); a site now termed the 'allosteric drug and metabolic site' (ADaM) (Langendorf & Kemp, 2015).Longchain fatty acylCo-As with a length between 12 and 18 carbon units, such as palmitoylCo-A have recently been identified as possible endogenous ligands for the ADaM site but are only active at β1-containing heterotrimers (Pinkosky et al., 2020).
Dysregulation of its activity is associated with ageing and neurological disorders although the precise mechanisms by which it becomes dysregulated and how changing its activity can influence disease course remain to be elucidated (Ma et al., 2014;Mairet-Coello et al., 2013;Salminen et al., 2011;Wang, Pan, & Chen, 2012).
A number of small molecules have been identified that directly activate AMPK.A769662 (Cool et al., 2006) binds to the ADaM site, which is proximate to the carbohydrate-binding module and the kinase domain, stabilising interaction between the two and inhibiting dephosphorylation; this interaction is further stabilised by AMP binding to the γ subunit (Sanders et al., 2007).A769662 activates AMPK heterotrimers containing the β1 subunit with tenfold specificity over β2 (Scott et al., 2008).PF06409577 similarly is thought to bind to ADaM and inhibit dephosphorylation of Thr172.This activator is highly specific, with thousand-fold higher activity at β1-containing isoforms of AMPK than β2 (Cameron et al., 2016).
when metabolised by adenosine kinase produces a phosphorylated metabolite ZMP (Gadalla et al., 2004).ZMP mimics AMP and primes AMPK for phosphorylation and hence, activation (Corton et al., 1995).The use of AICAR also causes increased levels of nucleotide monophosphate, which can have multiple sites of action, making it difficult to divorce the effects of AMPK activation from those caused by increased nucleotide monophosphate levels (Hasenour et al., 2014).A number of investigations using an AMPK knockout mouse have illustrated AMPK-independent effects of AICAR, including activation of fructose 1,6 bisphosphatase (Du et al., 2019;Hasenour et al., 2014).AICAR has been shown to also activate AMPK in an LKB1-independent manner in a pathway involving the ataxia-telangiectasia mutated protein kinase, a member of the phosphatidylinositol 3-kinase superfamily (Sun et al., 2007).
Metformin, a compound in common use as an antidiabetic therapeutic, was originally suggested to activate AMPK through inhibition of complex 1 of the electron transport chain leading to increased AMP levels (Owen et al., 2000) and decreased production of free radicals (Batandier et al., 2006).The suggestion that metformin inhibits Complex 1 is reinforced by more recent evidence showing that metformin does this in the absence of AMPK α1 or β1 subunits (Kelly et al., 2015).Metformin also directly inhibits mitochondrial glycerophosphate dehydrogenase in hepatocytes, which results in decreased activity of the glycerophosphate shuttle and altered cytosolic redox (Madiraju et al., 2014) with measured inhibition at concentrations as low as 50 μmol/L.This effect has also been shown in brain (Li et al., 2019) where mice were treated with 2 mg/mL metformin in drinking water for at least 4 weeks.The glycerophosphate shuttle operates particularly in oligodendrocytes although there is also evidence that it operates in astrocytes and neurons (McKenna et al., 2006).
Evidence also suggests metformin promotes the assembly of the AMPK heterotrimeric complex, which in turn increases the phosphorylation of AMPK at Thr-172 by LKB1 and also decreases the dephosphorylation at Thr-172 by protein phosphatase 2C (Meng et al., 2015).This mechanism is supported by the observation that the presence of LKB1 is required for the activation of AMPK by metformin (Shaw et al., 2005).In addition to reducing the production of reactive oxygen species in brain (Correia et al., 2008) metformin has also been suggested to ease nicotine withdrawal (Brynildsen et al., 2018) and to promote neurogenesis (Wang, Gallagher, et al., 2012).
Whether or not metformin has beneficial roles to play in the brain is arguable, with some animal studies in models of epilepsy and neurodegenerative disorders showing positive outcomes (Garg et al., 2017;Li et al., 2012;Martin-Montalvo et al., 2013;Zhao et al., 2014) and others not (Barini et al., 2016).The evidence in human trials directed to cognitive impairment and Alzheimer's disease is similarly contrary (Imfeld et al., 2012;Luchsinger et al., 2016;Moore et al., 2013).It has been suggested that the effects are highly dose-dependent (Mostafa et al., 2016).
While the metabolic consequences of activating AMPK have been explored in tissues such as liver and muscle, less is known about its actions in brain, despite the above-noted involvement of AMPK dysregulation in disease processes.Work in cultured astrocytes incubated with high (10 mmol/L) concentrations of metformin showed decreased TCA cycle flux and a doubling of glucose consumption and lactate release; these effects were attributed to inhibition of complex I in the respiratory chain (Hohnholt et al., 2017).
Infusion of metformin (0.2-10 mmol/L) into mouse brain has been shown to elicit massive increases in lactate and decreases in glucose, as well as evidence that it inhibits complex 1 in brain mitochondria (Thinnes et al., 2021).Inhibition of astrocyte AMPK with 0.5 mmol/L AICAR has been shown to impact astrocytic glutamate handling and metabolism (Voss et al., 2015) with a more recent suggestion that 0.5 mmol/L AICAR also increases astrocytic anaplerosis via pyruvate carboxylase and increases astrocytic glutamate oxidation (Voss et al., 2020).AMPK null mice lacking both β1 and β2 subunits have been shown to survive but to have lower levels of metabolites associated with energy metabolism (Muraleedharan et al., 2020).
Here, to test the metabolic consequences of activating AMPK in brain, we took advantage of known patterns of substrate compartmentation following the metabolism of [1-13 C]D-glucose and [1,2-13 C]acetate.Glucose is the mandatory brain substrate, which is metabolised by all cell types.Acetate, on the other hand, is metabolised mostly by a glial compartment, producing glutamine but is also metabolised in GABAergic neurons (Andersen et al., 2017;Rowlands et al., 2017).Metabolism of acetate is highly regulated and compartmentalised in brain, possibly because of its impact on the acetylation status of metabolising enzymes (Hallows et al., 2006;Rowlands et al., 2017)

| MATERIAL S AND ME THODS
All experiments were conducted in accordance with the guidelines of the National Health and Medical Research Council of Australia and approved by UNSW Animal Care Ethics Committee (20_105B).Guinea pigs (Cavea porcellus, wild type, male, 400-800 g, obtained from Pipers Farm), were fed ad libitum on standard guinea pig/rabbit pellets, supplemented with fresh greens and hay roughage and were maintained on 12 h light/dark cycle.
Male guinea pigs were used because of limited housing space and breeding constraints at the supplier.This study followed previously published protocols (Rae et al., 2000;Rae & Balcar, 2014) and was not pre-registered.to around 4 g of tissue per flask or 1 g/100 mL, which is the recommended slice/buffer ratio to avoid lactate inhibition of metabolism (Griffin et al., 1999).
All experiments were conducted for 90 min with 5 mmol/L [1-13 C] D-glucose and 0.5 mmol/L [1,2-13 C]acetic acid (control).This allows around one and a half turns of the Krebs cycle in the cortical tissue slice (McIlwain & Bachelard, 1985).In addition to a control flask, each activator was tested at two different concentrations, at or near the EC 50 and at 10 times that concentration to allow for maximal and also non-specific effects.Metformin was tested at three different concentrations to better investigate the multiple reported effects of this compound (detailed above).Drugs were dissolved in K-H buffer rather than dimethylsulfoxide (DMSO) because of the noted effects of this vehicle on metabolism (Nasrallah et al., 2008).
The following experiments were conducted using a total of 20 guinea pigs (Figure 1): 1. Control, 0.8 μmol/L and 8 μmol/L A769662 using cortices from five guinea pigs.
At the end of the incubation period, slices were rapidly removed from the medium by vacuum filtration and frozen in liquid nitrogen for extraction (Figure 1).Frozen slices were pulverised and

| Statistical analysis
The experiment is designed as a population with repeated sampling.Data were analysed using IBM SPSS Statistics (v22; RRID:SCR_019096), as described previously (Achanta et al., 2017;Rowlands et al., 2015Rowlands et al., , 2017)).No blinding was used and no animals nor data were excluded.Experiments with AMPK activators and inhibitors were compared with control variables using a repeatedmeasures ANOVA and statistically significant variables identified post-hoc using the Scheffé's F test.Those variables that passed this multiple-comparison filter were then compared using the nonpara-

| Pattern recognition of the data
The overall patterns of metabolism resulting from the activation of AMPK were analysed using multivariate pattern recognition approaches.Data from experiments using glucose and acetate as substrate and AMPK activators were imported into SIMCA P+ (v11.5; Umetrics; RRID:SCR_014688) with each variable expressed as a ratio relative to the control mean (N = 4).This was done so that the analysis would reveal the differences between each experimental condition rather than the effect versus the control condition, which would otherwise contribute a large proportion of the variance in the data.Data were subjected to unit variance scaling to ensure that each variable contributed equally to the model.A goodness of fit algorithm was used to assess the robustness of the resultant model (generally accepted as R 2 > 0.6), which is then cross-validated by calculating a predicted residual sum of squares (Q 2 ) (Maher et al., 2013).
from [1-13 C]glucose and a decrease in net flux into citrate from [1,2-13 C]acetate.and Ala C3, as well as into GABA C2, Gln C4, Asp C2 and C3 and citrate C2 (Figure 6a).This increase was largely due to increased net flux from [1-13 C]glucose (Figure 6b) although there was also a small increase in the net flux into glutamine from [1,2-13 C]acetate (Figure 6c).

| Effect of
Direct AMPK activator PF06409577 at high concentration (70 nmol/L) also stimulated net flux, but the impact on glycolytic byproducts lactate and Ala was muted by comparison with 8.0 μmol/L A769662.There were large increases in net flux into Krebs cycle by-products including Glu C2 and C4, Asp C2 and C3 and citrate (Figure 6d).This stimulus operated on both substrates (Figure 6e,f).

| Pattern recognition of the data
We tested to see whether the metabolic response to an activator could be used to objectively define membership of drug class (70 nmol/L), with these also being each distinct from the low concentrations of each of these activators (Figure 7b).

| DISCUSS ION
Here, in contrast to previous work in brain which used relatively high concentrations of AMPK activators (Blumrich & Dringen, 2019;Hohnholt et al., 2017;Voss et al., 2015Voss et al., , 2020)), we used concentrations of activators equal to reported EC 50 values, as well as 10 times that amount, to detect potential non-specific effects.We have also used the more specific AMPK activators A769662 and PF06409577 whose impact on brain metabolism has not yet been reported.A common metabolic factor of all activators, induction of glycolysis with the production of lactate and stimulation of metabolic pools sizes was observed, but there were also distinct metabolic outcomes specific to each direct or indirect class of activators, as well as outcomes related to the concentrations of activators used (Figure 7).
A769662 has been reported to have off-target effects, but these have only been reported at relatively high (mM) concentrations of the activator (Cameron & Kurumbail, 2016;Moreno-Sánchez et al., 1999;Scott et al., 2014;Treebak et al., 2009).While specific for AMPK both direct activators are more potent at β1 containing receptors than β2, with A769662 being tenfold more potent (Scott et al., 2008) and PF06409577 one thousand-fold (Cameron et al., 2016).The concentration of AMPK activator used has previously been reported to impact metabolic outcomes with metformin, in particular, reported to have strong concentration-dependent effects (Rena et al., 2017).Here, the low concentrations of each activator resulted in either no effect on pool sizes (10 μmol/L AICAR) or reduced production of lactate, suggesting decreased overall metabolic activity.
This was more apparent in the metabolic patterns produced by the allosteric (direct) activators A769662 and PF06409577, which also resulted in generally decreased pool sizes (Figure 2f,h).
The effect of the two lower concentrations of metformin (0.1 and 0.5 mmol/L) is instructional in this case with the lower of the two producing a decrease in net flux into lactate C3 and increased net flux into isotopomers of glutamine, aspartate and citrate (Figure 3d), which was contributed to by changes in flux from both glucose and acetate substrates (Figure 3e,f).These metabolites (glutamine, aspartate and citrate) are associated with glial metabolism (Griffin et al., 1998;Norenberg & Martinez-Hernandez, 1979;Westergaard et al., 1994).The higher of the two low concentrations of metformin Metformin is known to inhibit mitochondrial respiration and gluconeogenesis in a time-dependent manner (Owen et al., 2000); brain does not operate gluconeogenesis, but inhibition of respiration by metformin is estimated to occur at concentrations as low as 100 μmol/L because of the accumulation of positively charged metformin by mitochondria in a time-dependent manner (Owen et al., 2000).Here, in the incubation time of 90 min, some accumulation would be expected to occur with a concomitant inhibition of  (Sibson et al., 1998).
Along with the little impact on net flux into glutamate, there was an increase in Gln C4 labelling from glucose along with decreased incorporation into isotopomers of aspartate and citrate.This pattern was also largely mirrored by 2 mmol/L metformin (Figure 4).Pyruvate clearance, whereby the term refers to the clearance of pyruvate via metabolism through any of its metabolising enzymes, is generally dominated by the activity of pyruvate dehydrogenase, since this is a far from equilibrium enzyme.It is a useful way to think about lactate production via lactate dehydrogenase since the nearequilibrium nature of lactate dehydrogenase and its near-zero flux control coefficient mean that lactate production will usually reflect the amount of pyruvate present (Fell, 1997).Rapid clearance of pyruvate will therefore be reflected in lower lactate concentrations.
The different effects of the high concentrations of indirect and direct activators can also be seen in the principal components plot in behavioural outcomes, such as the response to pain, for example (Inyang et al., 2019) indicating that the mode of activation of the enzyme and its outcomes has functional consequences.
As mentioned above, A769662 and PF06409577 are more potent at β1-than β2-containing AMPK isoforms with PF06409577 in particular being highly specific for β1 over β2.Here, we endeavoured to use these activators at similar relative concentrations (EC 50 ) and it is notable that PF06409577 produced a quite distinct metabolic profile, being the sole activator at the higher concentration to increase net flux into glutamate isotopomers (Figures 4 and 6) along with decreased net flux into Ala C3, implying that pyruvate clearance is not reduced and that Krebs cycle activity is not impaired.
High levels of β2 versus β1 subunit expression have been reported in neurons in embryonic rat brain, but the expression of β2 was reported to decrease to almost undetectable levels in adult rat brain tissues (Culmsee et al., 2001).This report differs somewhat from the reported expression in mouse brain (Turnley et al., 1999) where expression of both β subunits has been reported to be high (Dasgupta et al., 2012).Subunit expression in the guinea pig brain has not been reported to our knowledge.
A769662 has been shown to inhibit the Na + /K + ATPase in skeletal muscle (Benziane et al., 2009).The reported IC 50 for this (57 μM) is 5-to 6-fold higher than the concentration of A769662 used here (10 μmol/L) so is likely to have a relatively small effect on the observed metabolic outcome here.

| Limitations
The metabolic rate of brain cortical tissue slices is low compared with the situation in vivo (McIlwain & Bachelard, 1985) although in this work this is partially compensated for by an increased experimental time so that label can pass about one and a half turns of the Krebs cycle.The slice preparation also avoids consequences associated with blood-brain barrier permeability and peripheral metabolism; some activators, for example, are metabolised in liver to form active catabolites (Ryder et al., 2018).We were also not able to examine possible sex effects as we explored the effects of AMPK activation solely in male guinea pigs because of housing space constraints.Sex may have some impact on the metabolic outcome of AMPK activation since upstream factors such as LKB1 are known to be under regulation by sex hormones (McInnes et al., 2012).

| CON CLUS ION
The metabolic outcomes resulting from the activation of AMPK are dependent on the degree of activation, and range from decreased

2. 1
| NMR spectroscopyAll spectra were acquired using a Bruker Avance III HD 600 spectrometer (Bruker Biospin) fitted with a cryoprobe (TCI; 5 mm) and refrigerated sample changer. 1 H spectra were acquired both with and without 13 C decoupling using adiabatic bilevel composite pulse decoupling across an effective bandwidth of 48 000 Hz during the acquisition time on a 30 s duty cycle to allow full longitudinal relaxation of resonances. 13C[ 1 H-Decoupled] spectra were acquired on a 4 s duty cycle using continuous WALTZ-65 decoupling.Total metabolite pool sizes (i.e. total of each metabolite whether labelled or unlabelled) were determined using Topspin (v.3.5: RRID:SCR_014227) from integrals in the fully-relaxed 1 H[ 13 C-decoupled] spectra, and the concentration of 13 C-labelled compounds from 13 C[ 1 Hdecoupled] spectra following adjustment for relaxation and nuclear Overhauser effect relative to a previously acquired fully-relaxed 13 C spectrum.The internal intensity reference for all spectra was sodium [ 13 C]formate, which is freely resolvable from all other peaks in both the 1 H and 13 C spectra.Incorporation of label from [1-13 C] D-glucose and [1,2-13 C]acetic acid is as described in earlier publications(Das et al., 2020;Rowlands et al., 2015Rowlands et al., , 2017)).Metabolite pool sizes were calculated as μmol/100 mg protein while net flux values are μmol/100 mg protein/90 min.F I G U R E 1 Graphical timeline of experiments.
metric Mann-Whitney test (with N = 4 it was assumed data were not normally distributed) to compute a p-value and were considered significant at p < 0.05.Only variables passing both these two tests were reported as significantly different.All significance values quoted in the text are those generated by the Mann-Whitney U test comparing the median values of variables that were previously been found to also be significant using the ANOVA and Scheffé's F test.All are N = 4 with no missing values or outliers.Given the typical relative standard deviations in the values obtained for our NMR-sourced variables (2%-10%), this approach does not yield a statistically significant outcome for variables with low to medium values of Cohen's d (<1) indicating that only those differences with high effect size are rated as statistically significant.For comparative purposes as done previously when comparing multiple experiments with similar activating ligands(Furlong et al., 2016;Nasrallah et al., 2007Nasrallah et al., , 2010Nasrallah et al., , 2011;; Rae et al., 2014), data are shown as change relative to the control mean for each experiment while the statistical indicators shown on the graphs, prepared in GraphPad Prism (RRID:SCR_002798) are calculated based on the raw data.
the low concentration direct (A769662 0.8 μmol/L and PF06409577 7.0 nmol/L) activators, low concentration indirect activators (AICAR 10 μmol/L and Metformin 0.1 and 0.5 mmol/L), high concentration direct (A769662 8.0 μmol/L and PF06409577 70 nmol/L) activators and high concentration indirect activators (AICAR 100 μmol/L and Metformin 2.0 mmol/L).These were then compared using a Coomans plot(Coomans et al., 1984), which plots the class distances of two models against each other in a scatterplot.The distance to the model of an observation is the same as the residual standard deviation of the observation.The critical distance to the model is calculated from the F-distributed values of the residual standard deviation of each observation divided by the pooled residual standard deviation of the model and is set at the desired probability level (e.g.p = 0.05 or 0.01).By also plotting the critical distance for each model in the plot (red line), four diagnostic areas are created.The lower left corner is where prediction set samples that fit both models are found; the lower right and upper left areas are where samples that fit each particular model are found; and the upper right area contains observations that do not conform with either of the models (classes).3| RE SULTS3.1 | Effect of activators on metabolite pool sizesThe effect of the different AMPK activators on total measured metabolite levels in the slices is shown in Figure2.Low concentrations of indirect or direct activators either had no significant effect or acted to decrease the size of metabolite pools (Figure 2a,c,d,f,h), with direct (ADaM) activators A769662 (0.8 μmo/L) and PF06409577 (7.0 nmol/L) having the most impact on pool sizes (Figure 2f,h), and low concentrations of metformin (0.1 and 0.5 mmol/L) acting only to reduce the pool size of lactate (Figure 2c,d).By contrast, higher concentrations of all activators (including indirect activator AICAR, 100 μmol/L) produced robust increases in the pool sizes of all metabolites measured and also increased pool sizes relative to the lower concentration of that activator (Figure 2b,e,g,i).3.2 | Effect of indirect activators on net flux of 13 CAICAR at 10 μmol/L had no significant effects on net flux of either labelled substrate (Figure3a-c).Low concentrations of metformin (0.1 and 0.5 mmol/L) reduced net flux into lactate C3 (Figure3d,g).Metformin in the lowest concentration (0.1 mmol/L) also increased net flux into Gln C4, Asp C2 and C3 and citrate C2 (Figure 3d,e).While both substrates accounted for similar increases in flux into Gln and citrate, the increased flux into Asp isotopomers was mainly driven by flux from [1,2-13 C]acetate (Figure 3e,f).By contrast, metformin at 0.5 mmol/L resulted only in decreased flux into lactate C3 (Figure 3g,h) with no significant impact on net fluxes into other isotopomers or on net flux from [1,2-13 C]acetate (Figure 3i).Increasing the concentration of AICAR to 100 μmol/L produced large increases in net flux into lactate C3 and Ala C3 with concomitant increases in net flux into Gln C4 (Figure 4a,b) and decreased incorporation of carbon from glucose into Asp C3. Less carbon from [1,2-13 C]acetate was incorporated into citrate C2,1 (Figure 4c).Similarly, raising the concentration of metformin to 2.0 mmol/L resulted in large increases in net flux into lactate C3 and Ala C3, as well as GABA C2 and Gln C4 (Figure 4d) driven by carbon supplied from both substrates (Figure 4e,f).As with AICAR, metformin resulted in a decrease in net flux into Asp C3 F I G U R E 2 Effect of AMP-activated protein kinase activators on metabolic pool sizes.Effect of each activator (2a-i) is shown as % change relative to the control mean with standard deviations.Statistics are calculated on the raw data.Total pool sizes are defined as the total amount of resonance in the fully-relaxed [ 13 C-decoupled] 1 H NMR spectrum, so represent both labelled and unlabelled metabolite.N = 4 samples of the cortical population, which was derived from N = 4-6 guinea pigs depending on the experiment (see Section 2 for details).
ADaM site activators on net flux of 13 C Low (0.8 μmol/L) concentration A769662 resulted in decreased net flux of carbon into most isotopomers measured (Figure 5a) but increased net flux into Asp C2 and C3.Inspection of substrate-specific isotopomers indicated that the increase in Asp labelling was mostly due to increased flux into Asp from [1,2-13 C]acetate (Figure 5c).Citrate labelling from [1,2-13 C]acetate was also increased while labelling of citrate from [1-13 C]glucose was not changed (Figure 5b,c).PF06409577 (7.0 nmol/L) produced similar metabolic outcomes to 0.8 μmol/L A769662 (Figure 5d-f), with increased labelling of citrate from [1,2-13 C]acetate and decreased net flux into lactate C3 and alanine C3.Increasing the concentration of the direct AMPK activator A769662 to 8.0 μmol/L stimulated net flux into glycolytic by-products lactate C3 F I G U R E 3 Effect of low concentrations of indirect AMP-activated protein kinase activators 5-Aminoimidazole-4-carboxamide ribonucleoside (AICAR) and Metformin on net flux of 13 C from [1-13 C]glucose and/or [1,2-13 C]acetate.Effect of each activator is shown as % change relative to the control mean with standard deviations.Black graphs are total net flux (all isotopomers of each carbon), blue graphs show isotopomers derived from [1-13 C]glucose, red graphs show isotopomers derived from [1,2-13 C]acetate.Statistics are calculated on the raw data.N = 4 samples of the cortical population, which was derived from N = 4-6 guinea pigs depending on the experiment (see Section 2 for details).*p < 0.05 compared with control.

F I G U R E 4
Effect of higher concentrations of indirect AMP-activated protein kinase activators 5-Aminoimidazole-4-carboxamide ribonucleoside (AICAR) and Metformin on net flux of 13 C from [1-13 C]glucose and/or [1,2-13 C]acetate.Effect of each activator is shown as % change relative to the control mean with standard deviations.Black graphs are total net flux (all isotopomers of each carbon), blue graphs show isotopomers derived from [1-13 C]glucose, red graphs show isotopomers derived from [1,2-13 C]acetate.Statistics are calculated on the raw data.N = 4 samples of the cortical population, which was derived from N = 4-6 guinea pigs depending on the experiment (see Section 2 for details).*p < 0.05 compared with control.# p < 0.05 compared with the lower concentration of activator.(activation site) and concentration (around Ka or higher).The low concentrations of indirect AMPK activators (AICAR 10 μmol/L, Metformin 0.1 and 0.5 mmol/L) were shown to produce metabolically distinct outcomes from low concentrations of the direct activators A769662 (0.8 μmol/L) and PF06409577 (7.0 nmol/L), and these metabolic outcomes were also distinct from those generated by high concentrations of each of these activators (Figure 7a).Similarly, high concentrations of indirect activators (AICAR 100 μmol/L, Metformin 2.0 mmol/L) generated statistically different metabolic patterns to high concentrations of A769662 (8.0 μmol/L) and PF06409577

F
Effect of low concentrations of direct AMP-activated protein kinase activators A769662 and PF06409577 on net flux of 13 C from [1-13 C]glucose and/or [1,2-13 C]acetate.Effect of each activator is shown as % change relative to the control mean with standard deviations.Black graphs are total net flux (all isotopomers of each carbon), blue graphs show isotopomers derived from [1-13 C]glucose, red graphs show isotopomers derived from [1,2-13 C]acetate.Statistics are calculated on the raw data.N = 4 samples of the cortical population, which was derived from N = 4-6 guinea pigs depending on the experiment (see Section 2 for details).*p < 0.05 compared with control.

( 0 .
5 mmol/L) by contrast appears to have very little impact on metabolism, with the main impact being a decrease in both the lactate pool and the net flux of carbon into lactate C3.Increasing the concentration of metformin to 2 mmol/L then produces strong metabolic effects (Figure4d) with increases in net flux into glycolytic by-products lactate C3 and Ala C3, as well as increased net flux into GABA and glutamine from both labelled substrates (Figure4e,f).By contrast, net flux into glutamate was not altered (Figure4d-f).Taken together, this suggests two separate effects of metformin; one that decreases metabolism and one that greatly increases it, with the crossover point occurring around 0.5 mmol/L concentration in this case.

F
Effect of higher concentrations of direct AMP-activated protein kinase activators A769662 and PF06409577 on net flux of 13 C from [1-13 C]glucose and/or [1,2-13 C]acetate.Effect of each activator is shown as % change relative to the control mean with standard deviations.Black graphs are total net flux (all isotopomers of each carbon), blue graphs show isotopomers derived from [1-13 C]glucose, red graphs show isotopomers derived from [1,2-13 C]acetate.Statistics are calculated on the raw data.N = 4 samples of the cortical population, which was derived from N = 4-6 guinea pigs depending on the experiment (see Section 2 for details).*p < 0.05 compared with control.# p < 0.05 compared with the lower concentration of activator.
contrast, has very little impact on metabolism at 10 μmol/L with no significant effects on metabolite pool sizes or net fluxes (Figures2a and 3a-c).Increasing the concentration to 100 μmol/L has strong effects, similar to 2 mmol/L metformin, as illustrated by the clustering of the two in the scatter plot showing the first two principal components of the model comprising all the data (Figure7a).This confirms previous reports of increased lactate production, with concomitant production of alanine once the limited ability of lactate dehydrogenase to buffer increased pyruvate levels is exceeded.Inspection of carbon spectra showed no visible peak from glycerol-3-phosphate suggesting that the redox balance (NAD + /NADH) is maintained within 'normal' levels by the buffering of pyruvate by alanine aminotransferase(Ben- Joseph et al., 1993).Voss et al. have previously shown an impact of 0.5 mmol/L AICAR on the metabolism of labelled glutamate with a reduction in the use of glutamate by astrocytes, no change in glutamate labelling and no impact on glutamate uptake.This is consistent with the lack of effect of 100 μmol/L AICAR on net flux into Glu isotopomers, which are generally representative of Krebs cycle flux

F
Low and high concentrations of indirect or direct AMP-activated protein kinase activators have distinct metabolic effects.(a) Scatter plot showing the first two principal components of a model created with SIMCA P+ where the change in concentration relative to the control mean for each metabolite pool and isotopomer was imported.Note that data for low indirect activators (0.5 mmol/L metformin) and high direct activators (8 μmol/L A769662) are separated in t[3] by at least three scores (data not shown) although this distance cannot be visualised when looking only at t[1] and t[2] as in this figure.The ellipse represents Hotelling's T 2 corresponding to a multivariate generalisation of the 95% confidence interval.Data from each experimental condition expressed relative to the control mean was imported into SIMCA P+ and subjected to principal components analysis.(b) Variable loadings contributing to the first principal component (t[1]).(c) Variable loadings contributing to the second principal component (t[2]).(d) Coomans plot (see Section 2 for explanation of Coomans plots) showing the distance between the models generated by data from low indirect and low direct (blue) activators.(e) Coomans plot showing the distance between the models generated by the high indirect (red) and high direct (green) activators.The grey circles represent the rest of the total dataset.4.1 | Distinct metabolic effects of direct and indirect activatorsWhile the direct activators A769662 and PF06409577 at high concentration also resulted in large increases in net flux from [1-13 C]glucose into lactate C3 (Figure6) they produced a different impact on the incorporation of 13 C into Krebs cycle by-products.This can be seen by the increased net flux into isotopomers of Gln, Asp and citrate (Figure6); this increase is more apparent with the more potent activator PF06409577, where it appears that pyruvate clearance is no longer impeded as net flux into Ala C3 is decreased (Figure6d,e).

(
Figure 7a) where the direct activators separate in the upper righthand quadrant of the Hotellings circle from the indirect activators in the lower right-hand quadrant.The loadings for principal component 2 (which separates the data in the vertical direction in the scatter plot) illustrate the metabolites that contribute most to this separation, while the Cooman's plot (Figure 7e) illustrates the statistical significance of the difference.A769662 has been shown to activate pyruvate dehydrogenase in a model of advanced breast cancer through phosphorylation of PDHA, where this activation maintains Krebs cycle function despite increased glycolytic activity (Warburg effect) in these cells(Cai et al., 2020).Whether this is also occurring in brain remains to be demonstrated.While at the lower concentrations, the effects of the direct and indirect activators were similar, they could still be differentiated by their metabolic patterns although the distance between the two models of low concentration direct and indirect activators was small (see Figure7a,d).The direct activators were distinguished particularly from the indirect activators by decreases in metabolic pool sizes in addition to lactate (Figure 2) and increases in net flux into citrate from [1,2-13 C]acetate (Figures 5c,f vs. 3c,i).While both indirect and direct activators increase AMPK phosphorylation, indirect activators also have addition effects such as the activation of upstream kinases such as LKB1 and impact the balance of ATP/ADP, while the direct activators would increase AMPK activity without necessarily altering cellular phosphorylation potential.Distinct differences in the mode of AMPK activation have been demonstrated metabolic pool sizes at very low activation to increased aerobic glycolysis and decreased pyruvate clearance at higher activation.Specific activation of β2-containing AMPK isoforms may also result in increased Krebs cycle activity, restoring pyruvate clearance while also increasing the rate of lactate production.AUTH O R CO NTR I B UTI O N S Conceptualisation: Caroline D. Rae.Methodology: Caroline D. Rae, Donald S. Thomas and Lavanya B. Achanta.Validation: Lavanya B.