Metformin causes a futile intestinal–hepatic cycle which increases energy expenditure and slows down development of a type 2 diabetes-like state

Objective Metformin, the first line drug for treatment of type 2 diabetes, suppresses hepatic gluconeogenesis and reduces body weight in patients, the latter by an unknown mechanism. Methods Mice on a high fat diet were continuously fed metformin in a therapeutically relevant dose, mimicking a retarded formulation. Results Feeding metformin in pharmacologically relevant doses to mice on a high fat diet normalized HbA1c levels and ameliorated glucose tolerance, as expected, but also considerably slowed down weight gain. This was due to increased energy expenditure, since food intake was unchanged and locomotor activity was even decreased. Metformin caused lactate accumulation in the intestinal wall and in portal venous blood but not in peripheral blood or the liver. Increased conversion of glucose-1-13C to glucose-1,6-13C under metformin strongly supports a futile cycle of lactic acid production in the intestinal wall, and usage of the produced lactate for gluconeogenesis in liver. Conclusions The reported glucose–lactate–glucose cycle is a highly energy consuming process, explaining the beneficial effects of metformin given continuously on the development of a type 2 diabetic-like state in our mice.


INTRODUCTION
Type 2 Diabetes Mellitus (T2DM) is one of the most common disorders in industrialized countries, with rapidly increasing patient numbers in the last decades; thus, its successful treatment in the setting of a metabolic syndrome gets more and more important. According to the 2015 guidelines of the ADA (American Diabetes Association) and the EASD (European Diabetes Association), lifestyle modification, i.e. weight control and physical activity, in combination with metformin (1,1-dimethylbiguanide) is the current first-line therapeutic concept for T2DM patients [1,2]. Basically, metformin lowers elevated blood glucose levels, and, with the successful treatment of hyperglycemia, it results in a significantly reduced diabetes-related morbidity [3]. Beyond its glucose lowering effect, metformin-treatment results in significant weight loss (summarized in recent large meta-analyses [4,5]). Surprisingly, the mechanisms responsible for lowering body weight are unknown, even though weight loss alone improves glucose homeostasis in T2DM. Although being introduced and available for clinical use since the 1950ies, metformin's therapeutic mechanisms are still not understood. One of the earliest possible modes of action, then identified for alkylguanidines that are closely related to the biguanidine metformin, was inhibition of oxygen consumption in liver mitochondria [6]. Later, it was shown in isolated hepatocytes that metformin in high 1 concentrations specifically inhibits complex I of the respiratory chain and reduces gluconeogenesis, probably by inhibiting pyruvate carboxylase inside mitochondria, a rate limiting enzyme of this process which is sensitive to the cellular energy as well as the redox state [7,8]. Stimulated by the finding that the ATP/ADP ratio was lowered by metformin in liver, the AMP-activated protein kinase (AMPK) was subsequently shown to be activated by metformin in isolated hepatocytes [9]. However, later it was shown that in mice lacking both isoforms of the AMPKa subunit as well as the upstream kinase LKB1, the hypoglycemic effect of the drug was still maintained [10,11]. Since then, new alternative mechanisms explaining how metformin may inhibit hepatic gluconeogenesis have been proposed, e.g. suppression of glucagon signaling by interfering with cAMP production [12], altering the hepatic redox state by direct inhibition of mitochondrial glycerophosphate dehydrogenase [13] as well as activation of a neurohumoral gutebraineliver axis [14]. Most studies still concentrate on the liver as the main target of metformin, arguing that this organ plays the key-role in gluconeogenesis and that intracellular drug concentrations will reach high levels after orally administered metformin is absorbed by the intestine [15]. It was previously shown that an important mode-of-action of the drug is to improve lipid homeostasis by stimulating the AMPK mediated phosphorylation of acetyl-CoA carboxylases, which consequently improves insulin sensitivity [16]. Although elegant, it is important to note that the drug was applied by intraperitoneal injection, thus bypassing the physiological route of orally taken metformin. However, after oral administration, the highest concentration of metformin is not found in the liver but in the intestinal epithelium [17e19]. Early data using obese fa/fa rats already showed that metformin administration significantly increased glucose consumption in the intestine [20] due to mitochondrial inhibition, which was confirmed later demonstrating increased lactate production in isolated human jejunal preparations [19]. Here, we continuously fed metformin in a therapeutically relevant dose, mimicking a retarded formulation, to mice on a high fat diet in order to investigate how the drug slows down the development of T2DM, but most importantly, how it slows down weight gain, the other well described mode of action in patients.

Animals and experimental protocols
Animals (male C57BL6/J mice) were housed in a 12 h light-dark circle (06:00 on, 18:00 off, including a period of dawn) at constant temperature of 22 C and a humidity of 60 rH. Controls received standard chow (Altromin Spezialfutter, Lage, Germany, #TPF-1314: 5% fat, 4.8% disaccharide, 23% protein). HFD treated mice were fed a high fat, high sucrose diet (HFD; Altromin Spezialfutter, Lage, Germany, #105712: 35% fat, 19% disaccharide, 19% protein). HFD þ Met and HFD þ lateMet received the same HFD supplemented with 0,5% of metformin (1,1-Dimethylbiguanide Hydrochloride, 97%, Sigma Aldrich, Darmstadt, Germany, #D150959). Metformin was added to the diet during the manufacturing process to assure equal distribution over the HFD. Mice had access to chow and water ad libitum unless otherwise specified. A detailed description about the different cohorts of mice used in this study can be found in the supplementary experimental procedures. All animal procedures were performed in accordance with the German Laws for Animal Protection and were approved by the local animal care committee (Landesamt für Natur-, Umwelt und Verbraucherschutz, LANUV, Recklinghausen, Germany; Az 37.09.298).

Metabolic characterization and glucose tolerance test
Mice were weighed every two weeks, starting at 6 weeks of age. Intraperitoneal glucose tolerance tests (GTT) were performed every 2 weeks, starting at 12 weeks of age, after 16 h of fasting by an intraperitoneal injection of 2 g/kg glucose (Supplementary Figure 1). Glucose was measured in tail venous blood at 15, 30, 60 and 120 min after injection (Glucomen LX, Berlin-Chemie, Berlin, Germany).
2.3. Indirect calorimetry and physical activity measurement At the age of 6 weeks, this cohort of animals was divided into two groups and fed HFD þ Met or HFD (n ¼ 16 each) for 12 weeks. At the age of 18 weeks, mice were kept for 48 h (starting at 10 am) in an open circuit measurement system (PhenoMaster, TSE Systems GmbH, Bad Homburg, Germany) after having been acquainted to the new environment. Light phase was from 7:30 PM until 6:30 AM. The following parameters were obtained: CO 2 production, O 2 consumption, home cage activity, food intake, water intake and feeding events, and values for energy expenditure (EE) and respiratory exchange ratio (RER; CO 2 production/O 2 consumption) were derived from these measurements.
2.4. Inhibition of mitochondrial complex I and integrity of the intestinal mucosa in the presence of metformin Mice were sacrificed by cervical dislocation; the small intestine (duodenum) was removed and immediately cooled on a metal plate kept on ice to slow down self-digestion as much as possible. Samples from the duodenum were immersed in Tissue Tek (O.C.T.Ô Compound e Sakura Finetek, Staufen, Germany), immediately frozen in liquid nitrogen, and sectioned using a Leica CM1950 cryostat (Leica Microsystems, Wetzlar, Germany) at À20 C. In order to quantitate NADH dehydrogenase (Complex I) activity, the following method was used (Diaphorase activity, modified from [21]: directly after preparation, 2e 3 mm sections were incubated in 0.1 M Tris, pH 7.4, 4 mM NADH, 0.1 M TriseHCl, pH 7.4, 0.2 mM nitroblue tetrazolium chloride for 10 min at room temperature. Sections were then briefly rinsed with distilled water, dried, and mounted in aqueous medium. Sections incubated without NADH were used as controls. Before, it was established that after 10 min, staining had not reached maximal intensity, thus allowing determination of Complex I activity in situ in different samples processed in a highly parallel way. Images were taken at 20Â or 40Â magnification using an Olympus BX-40 microscope (Olympus, Hamburg, Germany) and staining intensity was analyzed using a thresholding tool-based method (Using ImageJ [22], and the image processing package Fiji for Image J [23,24].). The thresholding tool settings were established in samples from HFD mice and used for quantification of all samples. The ratio of staining-positive to total area of 2e4 samples of the duodenum from HFD and HFD þ Met mice (n ¼ 2) were determined. Staining procedures for hematoxylin-eosin (HE) and PCNA (Proliferating cell nuclear antigen) are described in the supplementary experimental procedures. Waters, Eschborn, Germany), A 10 mM ammonium formate pH 3.8, B methanol, gradient flow 0.4 ml/min, 80% B at 0 min, 20% B at 2 min, 20% B at 4 min, 80% B at 6 min, stop at 8 min. Atmospheric pressure ionization with positive or negative electrospray was used. The following fragments were chosen for selected reaction monitoring (m/z parent, m/z fragment, collision energy (V; minus indicates negative ion detection)): glucose, 179, 89, À10; glucose-1-13 C, 180, 90, À12; glucose-1,6-13 C, 181, 90, À30; lactate, 89, 43, À20; lactate-3-13 C, 90, 44, À13; metformin, 130, 60, 19; methionine, 150, 133, 15. For each analyte, the peak area was obtained from integrating intensity above background vs. time in the proper elution time interval. Methionine was chosen to normalize the cheek punch data because here the Figure 1: Metformin slowed down the development of obesity and type 2 diabetes in mice upon a high-fat diet and increased energy expenditure. (A þ B) Compared to mice fed with high fat diet (HFD; n ¼ 8), metformin (Met) treatment significantly slowed down weight gain in mice fed with high fat diet and treated with metformin starting at the same time (HFD þ Met; n ¼ 11; p < 0.01) and HFD mice treated with metformin after already developing glucose intolerance (HFD þ lateMet; n ¼ 8; p < 0.01) mice. Control mice (n ¼ 11) received standard chow and their weight gain was considered as normal (p < 0.01 vs. HFD and HFD þ lateMet, respectively). (C þ D) Indirect calorimetry and physical activity measurements after 12 weeks of treatment showed that (C) energy expenditure was persistently higher in HFD þ Met (n ¼ 16) mice, whereas (D) their locomotor activity was significantly lower during dark phases compared to HFD mice (n ¼ 16). For statistical analysis of energy expenditure and activity, see Supplementary Figure S2. (E) Results of glucose tolerance tests (GTTs) at 18 weeks treatment are shown as areas under the curve. There were no differences in glucose tolerance between control (n ¼ 10) and HFD þ Met (n ¼ 11) mice (p ¼ n.s.). Glucose tolerance of HFD (n ¼ 7) and HFD þ lateMet (n ¼ 8) was significantly higher compared to control mice (p < 0.05). (F) Compared to HFD (n ¼ 7) mice, glycemic control, judged by glycated hemoglobin A1c (HbA1c) levels, was significantly improved in control (p < 0.05; n ¼ 9), HFD þ Met (p < 0.01; n ¼ 9) and HFD þ lateMet (p < 0.05; n ¼ 8) mice. There were no differences between HFD þ Met, HFD þ lateMet and control mice (p ¼ n.s.). Data are expressed as mean AE SD; for clarity, SDs are not shown in Figure 1 (C) and (D). n indicates the number of analyzed mice or of individual mice from which blood was collected and analyzed. Differences between groups were analyzed by 1-way ANOVA, followed by Bonferroni's post-hoc test; *p < 0.05 and **p < 0.01. peak area followed dilution steps best. However, similar results were obtained with leucine, phenylalanine, and valine. Serum data were not normalized.

Further experimental procedures
Procedures describing immunoblotting, qRT-PCR and determination of energy content of feces are described in detail in the supplementary experimental procedures. In short, pACC/ACC ratios and quantification of Ucp1 expression in BAT and WAT were determined by immunoblotting of proteins from liver samples after 12 weeks of treatment. Expression of monocarboxylate transporters and lactate dehydrogenase isoforms were analyzed by qRT-PCR. Analysis of the energy content of feces was performed by Fourier-transformed infrared (FT/IR) reflectometry and following bomb calorimetry.

Statistical analysis
All data are presented as means AE S.D. Areas under the curve (AUC) were calculated using the trapezoidal method [25]. Differences between groups were analyzed by Student's t-test or 1-way ANOVA, followed by Bonferroni correction, as appropriate (see Figure legends). Statistical significance of post-hoc analyses was accepted at p < 0.05. Calculations were performed using SPSS21 (IBM Deutschland, Ehningen, Germany).

Continuous metformin application slows down body weight gain
Based on an extended pre-experiment, four groups were established (Supplementary Figure 1). In one group, metformin was given to mice that had already developed glucose intolerance (HFD þ lateMet); the other group was treated together with the onset of HFD feeding (HFD þ Met). Considering the food intake of our mice (about 0.08 g g À1 body weight per day of the HFD þ 0,5% metformin diet), daily metformin intake was about 500 mg kg À1 per day, which equates to a human equivalent dose (HED) [26] of about 40 mg metformin kg À1 per day. Extrapolated to the average patient's weight, this dose is in the same range as the maximum dose of metformin given to humans (3 Â 1000 mg per day) [27]. Weight development of the four groups is shown in Figure 1A. For better illustration and quantitation, we compared the increase of weight between 6 and 18 weeks ( Figure 1B). While the HFD group gained 9.6 g, metformin considerably reduced weight gain, with only 5.8 g weight gain in the HFD þ lateMet group and even only 3.6 g in the HFD þ Met group (p < 0.01 vs. HFD, respectively).

Continuous metformin application increases energy expenditure
In order to find out how metformin treatment results in reduced weight gain, which is probably equivalent to weight loss in patients starting treatment and equivalent to improved maintenance of weight during treatment [28,29], whole body metabolism was analyzed at the age of 18 weeks, after 12 weeks of HFD vs. HFD þ Met. Reduced weight gain cannot be explained by decreased food intake, since this was similar in both groups, even though mice consuming metformin weighed less (HFD: 0.08 AE 0.03 g/day Â g body weight vs. HFD þ Met: 0.09 AE 0.02 g/day Â g body weight; p ¼ 0.29; Supplementary  Figure 2A). However, energy expenditure of the HFD þ Met mice was significantly higher, no matter if calculated from data for the entire 48 h period ( Figure 1C; HFD: 21.1 AE 4.0 kcal/h Â kg body weight vs. HFD þ Met: 22.5 AE 4.9 kcal/h Â kg body weight; p < 0.001) or by separating it into light and dark phases (Supplementary Figure 2B). Food, and thus also metformin, was consumed mostly during the dark phase (HFD: 18.13 AE 2.97 vs. HFD þ Met: 19.13 AE 2.42 feeding events; p ¼ n.s.), but also quite regularly during the light phase (HFD: 9.5 AE 3.3 vs. HFD þ Met: 8.75 AE 2.3 feeding events; p ¼ n.s.). Higher energy expenditure cannot be explained by increased locomotor activity, since HFD þ Met mice were even less active ( Figure 1D; HFD: 1524 AE 2182 counts/30 min vs. HFD þ Met: 812 AE 688 counts; p < 0.001), mainly caused by a higher activity of the HFD group during the dark phase (Supplementary Figure 2C). Also the respiratory exchange ratio (RER: CO 2 production vs. O 2 consumption) was slightly higher in the HFD þ Met group (HFD: 0.821 AE 0.03 vs. HFD þ Met: 0.826 AE 0.03; p < 0.001). When the light and dark phases were analyzed separately, the higher RER of the HFD þ Met group was only seen in the dark phase, when animals are active and consume most of the food together with metformin (Supplementary Figure 2D). Water intake was significantly increased in the HFD þ Met group with 0.11 AE 0.03 ml/day Â g body weight vs. 0.08 AE 0.03 ml/day Â g body weight in the HFD group (p < 0.01). Since both groups had the same food intake, we asked if the reduction of weight gain might be a result of malabsorption due to an acute effect of metformin on mitochondrial performance in the intestinal transport epithelium. However, no significant differences were observed in energy content of feces between groups (HFD: 19.35 AE 1.04 kJ/g dry weight vs. HFD þ Met: 19.96 AE 0.52, p ¼ 0.23). We also measured total feces production, but found no differences (data not shown).

Continuous metformin application improves glycemic control and lipid metabolism
Intraperitoneal glucose tolerance tests (GTTs) were performed in all groups every two weeks starting at 12 weeks, i.e. 6 weeks after onset of HFD, and blood glucose values during the GTTs done are shown in Supplementary Figure 3. Glucose tolerance was impaired in the HFD groups starting at 14 weeks, resulting in higher glucose levels at each time point compared to normal chow controls. After 18 weeks, glucose tolerance was not different from controls in the HFD þ Met group, shown as areas under the curve ( Figure 1E). Fasted glucose levels were significantly, but only moderately elevated in the HFD groups (Supplementary Table 1) confirming our previous finding that it takes about 6 months to develop severe T2DM using C57BL6/J mice and this food composition [30]. Metformin did not ameliorate fasting hyperglycemia compared to the HFD group (Supplementary Table 1), probably also because the animals had not ingested the drug for 16 h before blood sampling, and the effects on glucose tolerance seemed rather moderate. However in humans suffering from T2DM, the level of HbA1c is a much more reliable parameter for the long-term follow up of glycemic control [31]. Indeed, HFD significantly increased HbA1c levels, while both metformin treatment schemes completely normalized them, recapitulating its well established effect on glycemic control in humans ( Figure 1F and Supplementary In summary, in mice upon a high fat diet, feeding metformin slows down the development of a type 2 diabetes-like phenotype with profound effects on obesity and long-term glycemic control.

No activation of brown adipose tissue
The increase of energy expenditure could be due to recruitment by metformin of brown adipose tissue (BAT) or of beige fat cells [32] in white adipose tissue depots (WAT). Both would be characterized by increased levels of uncoupling protein 1 (Ucp1) as well as subunits of the mitochondrial respiratory chain [33], here analyzed by the representative core subunit IV of cytochrome c oxidase (CoxIV). However, no differences were found for Cox IV, neither in BAT nor in WAT, and levels of Ucp1 were unchanged in BAT, while it was not detectable at all in WAT (Supplementary Figure 4).

Evidence for a futile glucoseelactateeglucose cycle
Metformin has been shown to inhibit complex I of the mitochondrial respiratory chain in hepatocytes in high concentrations (100 mM up to 10 mM; [8]). In general, cells increase lactate production from glucose when mitochondrial ATP production is insufficient or inhibited and indeed, lactate levels rose in the intestinal wall, but not in liver, of mice which had been fed a HFD with metformin for 24 h (Figure 2A). Moreover, in blood taken from the portal vein of these mice, we found a significant increase in lactate and a significant drop of pH as well as of actual base excess (ABE), respectively (Table 1), clearly showing that metformin indeed causes lactic acid production in the intestinal wall, from which it is rapidly removed due to its high perfusion rate. In contrast, neither an increase in lactate nor evidence for acidosis was found in blood taken from the facial or tail vein, respectively (Supplementary Tables 2 and 3), indicating that lactic acid released into the portal circulation does not leave the enterohepatic vascular bed and therefore does not reach the general circulation. Since our results show that metformin indeed acts as a mitochondrial inhibitor in the wall of the small intestine, we stained samples of the duodenum for NADH-"diaphorase" activity, which is due to mitochondrial complex I activity in cells rich in mitochondria. This demonstrated the high abundance of this complex in the intestinal epithelium, both apical as well as basal ( Figure 3A), and showed that complex I activity was about 3 fold lower in metformin consuming mice ( Figure 3B). Staining for proliferating cells (PCNA; Figure 3C) and with H&E ( Figure 3D) showed no difference, demonstrating that metformin does not cause any tissue damage at the administered dose. Our results thus strongly suggest conversion of glucose to lactate in the wall of the small intestine in the presence of metformin due to complex I inhibition and the conversion of lactate back to glucose in the liver. Indeed, glycogen levels were considerably lower in livers of mice after 12 weeks of HFD þ Met, further corroborating this hypothesis (control: 355 AE 55; HFD: 301 AE 29 vs. HFD þ Met: 137 AE 39 mMol glycosyl units/g wet weight, p < 0.01 vs. HFD and vs. control, respectively). In order to demonstrate this pathway directly, mice were given labeled glucose-1-13 C in HFD. Since the natural abundance of the 13 C-isotope is only 1.1%, no doubly labeled glucose-1,6-13 C was detected in the blood samples before the experiment ( Figure 4A and Supplementary Figure 6). After 3 h of feeding, newly generated, doubly labeled  (Mct-1), a transporter specialized for lactate import, in heart (p < 0.05), muscle (p < 0.01), and liver (p < 0.01) and a trend for increased expression in the intestinal wall (p ¼ n.s.) of HFD þ Met (n ¼ 5) compared to HFD (n ¼ 6) mice. (C) Lactate levels were clearly reduced in chyme (p ¼ 0.054), intestinal wall (p ¼ 0.052), and liver (p < 0.05) of HFD þ Met (n ¼ 5) compared to HFD (n ¼ 6) mice after 3 weeks of treatment. (D) After 3 weeks of treatment, the ratio of lactate dehydrogenase (Ldh) isoforms Ldha/Ldhb were not different in heart and muscle (p ¼ n.s.), significantly decreased in liver (p < 0.01), and significantly increased in intestinal wall (p < 0.05) of HFD þ Met (n ¼ 5) compared to HFD (n ¼ 5) mice. Data are expressed as mean AE SD. n indicates the number of individual mice from which chyme and tissues were collected and analyzed. Differences between groups were analyzed by Student's t-test; *p < 0.05 and **p < 0.01. glucose-1,6-13 C was nearly two-fold higher in the presence of metformin compared to controls ( Figure 4A,B). Glucose-13 C, lactate-13 C, unlabeled glucose, and lactate were not significantly altered in peripheral blood ( Figure 4C,D). To even further support our findings (Figure 6), the expression of the monocarboxylate transporter isoform 1 as well as lactate dehydrogenase isoforms was analyzed by RT-qPCR in mice after 3 weeks of HFD, i.e. when we expected that processes adaptive to the constant release of lactate from the intestine had been fully initiated. In liver, heart and muscle, expression of the monocarboxylate transporter Mct-1 (Slc16a1), which is specialized for lactate import, was significantly decreased ( Figure 2B), while there was a trend for increased expression in the intestinal wall, in line with improved extraction of lactate from the chyme at that later stage ( Figure 2C). In addition, the ratio of lactate dehydrogenase (Ldh) isoforms Ldha/Ldhb was increased in the intestinal wall, favoring the conversion of pyruvate to lactate, while it was decreased in the liver, favoring the opposite reaction ( Figure 2D). Indeed, liver tissue showed significantly lower levels of lactate at this stage ( Figure 2C). Our results seem to be in conflict with a large body of literature showing inhibition of hepatic glucose production by metformin, although in the long term, we also found a significant improvement of glycemic control ( Figure 1E). Therefore, we analyzed phosphorylation of ACC at Ser 79, a well-known readout of metformin action in liver, in mice after 12 weeks of treatment, but found no significant changes compared to controls ( Figure 5A). However, applying metformin at a single high dose by gavage indeed led to increased ACC phosphorylation ( Figure 5B), as shown by others who used daily i.p. injections or gavage [16,34]. This demonstrates that at our feeding conditions, metformin obviously did not reach the liver in concentrations high enough to activate ACC phosphorylation and other downstream pathways. In order to substantiate this, metformin concentrations were analyzed and found to be 13 AE 5, 40 AE 13 and 37 AE 27 mmol/L after feeding HFD with 0.5%, 1% and 2.5% metformin, respectively. Thus, our feeding regime with 0.5% metformin resulted in levels which are in the lower range of concentrations reached by others in animals and humans (10e40 mmol/L) [34].

DISCUSSION
Numbers of prescriptions of metformin grew constantly in the last 20 years and it is now estimated to be given to at least 150 million people suffering from T2DM worldwide [34]. Since high blood glucose causes the most severe complications in T2DM, researchers interested in understanding its mode of action mostly concentrated on the suppression of glucose production in the liver, which certainly occurs in vivo in patients upon treatment [35]. In contrast, surprisingly little attention has been paid to the mechanisms responsible for lowering body weight upon or maintaining body weight during metformin treatment [4,5,28,29,35e37]. This effect cannot be explained by metformin's suppressive effect on hepatic glucose output, since gluconeogenesis is an energy consuming process; thus, metformin should decrease energy expenditure and therefore increase body weight. However, weight loss alone is a powerful means to improve glucose homeostasis in T2DM; therefore, we felt that deciphering the responsible mechanism also deserves attention. Indeed, in our study metformin not only slowed down the development of a type 2 diabetes-like phenotype with disruption of glycemic control and dyslipidemia, as expected, it also significantly slowed down weight gain. We first excluded the most trivial reasons like decreased food intake due to metformin's bitter taste, increased locomotion due to enhanced foraging activity, or malabsorption due to inhibition of active transport processes in the intestine. We also excluded activation of BAT or recruitment of beige adipocytes in WAT, respectively. Instead, the slower weight gain can be explained by the marked increase in energy expenditure we found after metformin treatment. We reasoned that metformin is an inhibitor of complex I of the respiratory chain, at least at high concentrations [7,38], which are certainly reached in the intestine after oral administration [18]. It has been reported before that metformin causes increased conversion of glucose to lactate in the intestinal wall, when given as single bolus at a dose about five times higher than the maximum daily dose given to patients ( [39], summarized in [19]). This treatment did not cause malabsorption of glucose, but the local lactic acidosis may explain the gastrointestinal disturbances reported for about 20e30% of patients taking metformin [11,40,41], some of which can even be associated with polymorphisms of putative metformin transporters [41]. Indeed, we found inhibition of complex I in the intestinal epithelium, followed by increased lactate levels and accompanied by a drop in pH in portal vein blood, but not in peripheral venous blood, and therefore conclude that lactic acid is converted back to glucose in the liver ( Figure 6). Most convincingly, we demonstrate that circulating levels of doubly labeled glucose-1,6-13 C were twice as high in the presence of metformin after oral administration of glucose-1-13 C ( Figure 4A,B). Breakdown of glucose-1-13 C will yield one molecule of lactate-3-13 C, which can be converted in the aldolase reaction into doubly-labeled glucose-1,6-13 C, when by chance two labeled lactate-3-13 C molecules are used for gluconeogenesis ( Figure 6). Since more lactate-3-13 C is generated in the intestinal wall in the presence of metformin (Figure 2A) and released into the portal vein (Table 1), it is more likely that doubly labeled glucose-1,6-13 C molecules are generated in the liver. Eventually, this process consumes 6 mol of ATP, when 1 mol of glucose is produced from 2 mol of lactate [42], which had generated 2 mol of ATP in the intestinal wall when derived from glucose, either taken from the lumen or from blood. We propose that this futile cycle, Table 1 e Increased blood lactate and decreased pH and ABE in blood from the portal vein in response to 24 h of metformin treatment. After 24 h of treatment, blood was analyzed from the portal vein of mice fed with high fat diet and treated with metformin (HFD þ Met; n ¼ 5), HFD mice (n ¼ 5) and control mice (n ¼ 6). In detail, pH, partial pressure of carbon dioxide (pCO 2 ), partial pressure of oxygen (pO 2 ), hemoglobin (Hb), oxygen saturation (sO 2 ), hematocrit, potassium, sodium, chloride, hydrogen carbonate (HCO 3 À ), actual base excess (ABE), glucose, and lactate were measured. Data are expressed as mean AE SD. n indicates the number of individual mice from which blood was collected and analyzed. Differences between groups were analyzed by 1-way ANOVA, followed by Bonferroni's post-hoc test; *p < 0.05 and **p < 0.01 HFD þ Met vs. control, # p < 0.05 HFD þ Met vs. HFD.  together with energy consuming mechanisms maintaining pHhomeostasis, largely explains the increased energy expenditure and, consequently, the reduced weight gain of metformin treated mice under a high fat diet. Indeed, liver glycogen content was significantly lower after metformin treatment, since additional resources are needed in the liver to support the energy consuming glucose production ( Figure 6, dotted line). In addition, our model is reinforced by the observation that after 3 weeks, several adaptive processes have occurred. In the intestinal chyme, the level of lactate was significantly lower in the metformin treated group, showing that the capacity to extract lactate from chyme has increased ( Figure 2C), probably by augmented expression of the responsible inward transporter Mct-1 ( [43]; Figure 2B; although not significant), and by a change in the ratio of Ldh isoforms [44] favoring production of lactate from pyruvate derived from absorbed or also blood-born glucose ( Figure 2D). Other metabolic organs have down-regulated Mct-1 ( Figure 2B), probably as a protection against inflow of lactate together with the accompanying proton. This may help to avoid intracellular acidosis in these organs when intestinal levels of metformin are high and lactic acid does appear in the periphery. In liver, the ratio of Ldh isoforms has changed favoring production of pyruvate from lactate ( Figure 2D) as a substrate for gluconeogenesis. Additional data supporting our model are an increased respiratory exchange ratio (RER) during the dark phase, when animals consume most of the drug (Supplementary Figure 3D), which is a well-established indicator of lactic acidosis, and a significantly increased water intake under metformin, which may be due to increased water loss by hyperventilation as a compensation for lactic acidosis.
Although we saw long-term improvement of glycemic control, based on HbA1c measurements and GTTs, we did not observe lower blood glucose in the fasted state, indicating that metformin did not importantly suppress hepatic gluconeogenesis in our mice. This seems to contradict an immense body of literature, both in patients [35] as well as in mice [12,13]. However, since phosphorylation of ACC was unchanged ( Figure 5A), we conclude that under the conditions of our experiment, metformin did not reach the liver in concentrations high enough to induce this downstream effect well described in previous studies. Indeed, 13 AE 5 mmol/L metformin which we obtained in peripheral blood with our feeding regime of 0.5% metformin is in the lowest range of concentrations reached by bolus administration in animals and humans reported by others [34]. When we administered the daily dose of metformin by gavage, as done by others, we also found increased ACC phosphorylation ( Figure 5B), emphasizing that hepatic metformin action is highly dose dependent, as discussed recently [34]. Patients take metformin (500e1000 mg) two or three times a day together with meals [39], and mice, in which phosphorylation of ACC has been shown in liver, are routinely given the drug by bolus gavage (15e500 mg/kg [12,45,46]) or daily i.p. injection (150 mg kg À1 [16]). In contrast, our mice consume 500 mg kg À1 distributed over 24 h, mostly, but not exclusively, during the night, thus other mechanisms due to different kinetics had to be expected. Metformin has been shown to activate anaerobic glycolysis in isolated intestinal preparations already at concentrations of 10 mM [47], probably due to the efficient translocation into enterocytes mediated by the apical transporters PMAT (SLC29A4) and SERT (SLC6A4) [48,49], but concentrations of 100 mM to even 10 mM have to be used in in peripheral blood, collected by cheek punch, and after 3 h of feeding; the ratio of doubly labeled glucose-1,6-13 C to methionine was significantly higher in mice fed with high fat diet and treated with metformin (HFD þ Met; n ¼ 4) compared to HFD mice (n ¼ 3; p < 0.05). (B) After 3 h of feeding, doubly-labeled glucose-1,6-13 C was significantly higher in sera, collected from the left ventricle, of HFD þ Met (n ¼ 4) compared to HFD mice (n ¼ 3; p < 0.05). (C) After 3 h of feeding, the amount of glucose and glucose-1-13 C and (D) lactate and lactate-3-13 C was not different in sera, collected from the left ventricle, of HFD þ Met (n ¼ 4) compared to HFD mice (n ¼ 3; p ¼ n.s.). Note, that because of variable blood sample size, the cheek punch data needed to be normalized, while data for serum generated from ventricular blood show absolute peak areas. Data are expressed as mean AE SD. n indicates the number of individual mice from which blood was collected and analyzed. Differences between groups were analyzed by Student's t-test; *p < 0.05.
hepatocytes to inhibit mitochondrial respiration there [38]. This is much higher than therapeutic concentrations and was definitely not reached in our setting. Therefore, we propose that the energy-consuming, futile glucosee lactateeglucose cycle, for which we have provided strong evidence here, will also operate in patients as long as the metformin concentration in the intestinal epithelium is high and low in the liver. Lactic acidosis remains mostly confined to the enterohepatic circulation, but may also affect other organs, which can be compensated by adaptive processes, e.g. changes of Mct-1 and Ldh isoform expression, as shown here in mice. The very rare cases in which patients experience systemic lactic acidosis are due to exceptionally high levels of metformin, e.g. because of inadequate excretion in patients with renal dysfunction [50,51] or polymorphisms in the organic cation transporters responsible for its route [41].
In patients taking their daily metformin dose, we propose two consecutive steps, as metformin will be first transported into the intestinal epithelium, which serves as a sink and where it inhibits the mitochondrial respiratory chain. It has been clearly shown that the drug is only slowly released into the portal vein [39], where it will inhibit gluconeogenesis in the liver only if concentrations are sufficiently high, which seems certainly to be the case in patients under normal treatment regimens [35]. Formulations of metformin with retarded release are already available on the market, and these may be adjusted to contain even higher concentrations in order to effectively drive both effects, increased energy expenditure due to the futile cycle described here as well as inhibition of gluconeogenesis, which will both ameliorate type 2 diabetes. Indeed, our work is strongly supported by a recent paper showing in large patient cohorts that even the glucose lowering effect of the drug may reside in the gut if given as a delayedrelease formulation (MetDR) [52]. To what extent the recently described effects of metformin on the composition of the gut microbiota contributes to our findings is unclear. However, it has been reported that in patients taking metformin, that there is a shift to  microbial species producing butyrate and propionate, metabolites which, in turn, stimulate intestinal gluconeogenesis [53], which in rodents results in a beneficial effect on glucose and energy homeostasis with reductions in hepatic glucose production, appetite and body weight [54] (reviewed in [55]).

AUTHOR CONTRIBUTIONS
The authors contributed to this work in the following ways: P.S.,