Blunted Brain Energy Consumption Relates to Insula Atrophy and Impaired Glucose Tolerance in Obesity

The incidence and prevalence of obesity have been escalating worldwide, and obesity has already reached epidemic proportions (1,2). This frightening development, in conjunction with the urgent need to replace more-or-less inefficient treatment strategies, has resulted in the development of new pathophysiological concepts of obesity. There is currently a growing consensus that this disease involves dysregulation within brain areas assigned to control food intake behavior and systemic energy homeostasis (3,4). Data show that complex neuronal pathways with reciprocal connections between the hypothalamus, brainstem, and higher cortical centers control appetite and food intake behavior (5), whereas afferent inputs from the periphery as well as efferent signals to peripheral organs regulate energy homeostasis (6). At large, appetite perception, food intake behavior, and energy homeostasis are synchronized in the hypothalamus as the cerebral “appetite center” (3,4). In this context, lower levels of high-energy phosphates (i.e., ATP and phosphocreatine [PCr]), have been detected in obese compared with normal-weight humans (7), which suggests a relationship between brain energy supply and body weight regulation. Supporting this view is the finding that cerebral ATP and PCr levels predict the amount of calories subsequently consumed (8). Moreover, brain energy consumption by electrical stimulation increases glucose tolerance in normal-weight men (9). On this basis, we aimed to test whether transcranial direct current stimulation (tDCS), a form of neurostimulation that uses constant low current delivered directly to the brain by small electrodes, would improve the characteristic glucose-intolerant state in obese volunteers. We further assumed that tDCS-induced energy consumption alters brain energy levels. Because obese individuals display not only reduced cerebral ATP and PCr levels but also lower gray matter density (10–12) and cortical metabolism (8) compared with normal-weight individuals, we likewise hypothesized that the morphometric decline could be due to the observed neuroenergetic deficit.

To test these hypotheses, we quantified brain energy levels and systemic glucose tolerance in obese versus normal-weight volunteers by ³¹P magnetic resonance spectroscopy (³¹P-MRS) before and after tDCS under the conditions of a standard hyperinsulinemic-euglycemic clamp procedure, which is considered the experimental gold standard to determine glucose tolerance in vivo (13). A glucose clamp places plasma glucose concentrations under the investigator’s control and, therefore, breaks the endogenous glucose-insulin feedback loop. This technique consists of an insulin infusion at a predetermined fixed dosage and a variable glucose infusion rate, which is continuously adapted to the target blood glucose. Under the steady-state conditions of euglycemia, the glucose infusion rate gives information about the systemic glucose tolerance of an individual.

In addition, we examined the voxel-based morphometry (VBM) measures of key brain areas assigned to appetite and taste processing in relation to systemic glucose uptake and brain energy content. VBM involves a voxel-wise comparison of the local volumes of gray matter between two groups of individuals (14). The measure of structural differences between populations derives from a comparison of the local composition of different brain tissue types (i.e., gray matter, white matter) (14). Overall, VBM has been assigned to be sensitive to these differences while discounting positional and other large-scale volumetric differences in gross anatomy.

Fifteen normal-weight (BMI 23.2 ± 0.38 kg/m²) and 15 obese right-handed men (BMI 36.3 ± 1.04 kg/m²) matched for age (24.6 ± 0.69 vs. 24.7 ± 0.66 years) participated in the experiments, and their metabolic characteristics are summarized in Table 1. All subjects had a regular sleep-wake cycle during the week before testing. Before participation, all volunteers completed the Structured Clinical Interview for DSM-IV Axis I Disorders (SCID-1). Volunteers with a psychiatric disease were excluded from the study. Further exclusion criteria were acute and chronic internal or neurological diseases, alcohol and drug abuse, smoking, shift work, competitive sports, exceptional physical or mental stress, and any kind of current medication. On the days before experimental testing, participants were instructed to go to bed no later than 2300 h and to avoid exhausting physical effort. The study was performed in accordance with the Declaration of Helsinki (2000) of the World Medical Association and approved by the University of Luebeck Ethics Committee. Each participant gave written informed consent.

The study was performed in a randomized sham-controlled crossover design. Each subject was tested in two experimental conditions spaced at least 2 weeks apart. On the days of experimental testing, subjects reported to the Department of Neuroradiology at 3:30 p.m. after fasting for at least 6 h. One cannula was inserted into a peripheral vein on the back of the hand and a second cannula into an antecubital vein of the contralateral arm. Baseline blood samples were collected for glucose and insulin measurements. The euglycemic-hyperglycemic clamp was started at 4:30 p.m. by administration of an bolus of H-insulin (Hoechst, Frankfurt, Germany) of 5 mU/kg body weight/min over 2 min. Afterward, insulin was infused at a constant rate of 1.5 mU/kg body weight/min until the end of the clamp. Simultaneously, a 20% dextrose solution was infused at a variable rate to maintain plasma glucose values at ∼5.5 mmol/L for 310 min. After euglycemic steady-state conditions were reached, structural magnetic resonance imaging (MRI) for VBM and baseline ³¹P-MRS were recorded (more details below). Subsequently, tDCS stimulation occurred, and a second sequence of ³¹P-MRS measurements was performed. In accordance with previous studies (7), a series of five continuous ³¹P-MRS sequences started 65 min and ended 105 min after tDCS. The last spectroscopy sequence was recorded after another 2.5 h to detect persistent tDCS-induced effects. Thereafter, the insulin infusion was stopped, and the plasma glucose concentration was normalized by increasing the glucose infusion rate.

For tDCS, two electrodes were applied. The anodal electrode was placed over the primary motor cortex representation of the left first interdigital muscle. Before placement, the “hot spot” of the left first interdigital muscle was identified using suprathreshold local transcranial magnetic stimulation in a stepwise manner by shifting the coil in 1-cm steps over the right vertex. The cathodal electrode was positioned over the left forehead. Stimulation electrodes were surrounded by a flat sponge soaked in an isotonic NaCl solution and fixed by elastic bands. tDCS was performed using a direct current-stimulator plus (neuroConn, GmbH, Illmenau, Germany), which delivered 20 min of anodal stimulation (1 mA, fade in/out 8 s). The procedure during sham stimulation was identical, except for the stimulator being turned off.

Structural MRI and cortical ³¹P-MRS were acquired in a 3.0-T MR scanner (Achieva 3T; Philips Medical Systems, Best, The Netherlands). Structural MRI was performed using a T1-weighted fast low angle shot three-dimensional MR sequence (echo time = 5 ms; repetition time = 15 ms; flip angle = 30°; isotropic voxel size = 1 × 1 × 1 mm³). Data were processed and examined using SPM8 software (Wellcome Department of Imaging Neuroscience, Institute of Neurology, UCL, London, U.K.) implemented in MATLAB 7.6 software (MathWorks, Sherborn, MA) and the VBM8 toolbox. Images were registered after applying a probabilistic framework using linear (12-parameter affine) and nonlinear transformations (warping). Subsequently, images were tissue-classified and bias-corrected within the same generative model (15) as well as tissue-classified into gray matter, white matter, and cerebrospinal fluid. Finally, modulated gray matter images were smoothed with a Gaussian kernel of 8 mm full width at half maximum.

The ³¹P-MRS images were acquired using a double-tuned ¹H/³¹P head coil (Advanced Imaging Research Inc., Cleveland, OH). Before acquisition, standardized volumes of interest were localized by taking scout images (Supplementary Fig. 1 illustrates the localization of the voxels). Overall, eight ³¹P-MRS sequences were measured as described in the study design. To allow satisfactory relaxation of the P metabolites, a repetition time of 4,500 ms combined with a three-dimensional chemical shift imaging sequence (6 × 5 × 3 voxel, 6 kHz bandwidth, 1,024 data points, 8:51 min measuring time; Supplementary Fig. 2 shows a representative spectrum) was chosen. For better spectral resolution during excitation and receiving, we applied ¹H decoupling and nuclear Overhauser effect (15) with a broadband proton decoupling (10 rectangular radio frequency pulses at a proton resonance frequency of 10 ms duration and 10 ms delay between them to generate a 90° flip angle on the ¹H nuclei) as well as ¹H decoupling (wideband alternating-phase technique for zero-residual splitting-4) (16) using the second channel of the head coil for transmitting on the ¹H resonance frequency. For evaluation of the spectral data, a magnetic resonance user interface was used. Zero filling to 4,096 data points and apodizing by a 20-Hz Lorentzian filter were applied. Peak positions and intensities were calculated using the advanced method for accurate, robust, and efficient spectral fitting (AMARES) algorithm (17). Before statistical analyses, spectral data were corrected for transmit power and receiver gains.

We examined the high-energy phosphate compounds ATP and PCr, which directly reflect the overall high-energy phosphate turnover (18). ATP was calculated as the sum of α-, β-, and γ-ATP. In addition to PCr and ATP, the ratios of PCr to inorganic phosphate (Pi) and ATP to Pi were evaluated as an indicator of intracellular energy status (7). High-energy compounds are presented as single values at each time point. These values depict an arithmetic mean of all data points measured over all voxels at a given time point.

Data are presented as mean values ± SEM. Statistical analysis was based on ANOVA for repeated measurements, including the factors “time” (time points of data collection), “treatment” (tDCS vs. sham), and “group” (for differences between normal-weight and obese participants), and the interaction effect between these factors. The unpaired Student t test was used for pairwise comparisons between groups and the paired Student t test was used to compare single time points between the conditions within one group. To account for baseline differences in the cerebral high-energy phosphate content between groups, data were transformed to relative changes from baseline and then included in the respective ANOVA models. Correlation analyses were performed by bivariate correlation analysis according to Pearson. Voxel-wise gray matter differences between obese individuals and normal-weight control subjects were examined using independent-sample t tests. To avoid possible edge effects around the border between gray and white matter or cerebrospinal fluid and absolute gray matter, the threshold of 0.01 (absolute threshold masking) was used. For statistical analyses, in the first step, we used an uncorrected threshold of P < 0.001 across the entire brain. In a second step, we performed multiple regression analyses (threshold of P < 0.001) to explore the association between regional brain volumes and spectroscopy values. Coordinates were reported in the standard anatomical space developed at the Montreal Neurological Institute. All tests comprised n = 15 in each group. A P value <0.05 was considered significant.

Plasma glucose concentrations were comparable between groups at baseline and throughout the euglycemic-hyperinsulinemic clamp (P = 0.460 for baseline plasma glucose, P = 0.520 for group main effect plasma glucose during clamp, P = 0.707 for group by time interaction). Insulin concentrations were significantly higher in obese than in normal-weight volunteers (P = 0.037 for baseline serum insulin, P < 0.001 for group main effect for serum insulin during clamp, P = 0.003 for group by time interaction; Fig. 1). Additional analyses per subject showed no differences in the insulin concentrations between single time points throughout the experiments in both groups (all P > 0.190).

Consistent with previous data, obese volunteers displayed distinctly reduced systemic glucose tolerance during the entire experiments compared with the normal-weight group (P < 0.001 for time main effect, P < 0.001 for group main effect, P < 0.001 for group by time interaction; Fig. 2). Accordingly, the overall glucose infusion rates required to maintain target blood glucose values during the clamp were significantly lower in obese compared with normal-weight participants (66.21 ± 3.95 vs. 85.27 ± 5.66 g; P = 0.010; Fig. 2, small insert).

Baseline cerebral PCr, ATP, and Pi content as well as PCr-to-Pi and ATP-to-Pi ratios in the two groups are presented in Fig. 3. Confirming previous data, one-way ANOVA showed significantly different PCr and ATP levels as well as PCr-to-Pi and ATP-to-Pi ratios, but no differences in the Pi values, between the two groups. Overall, obese men displayed lower cerebral PCr and ATP contents than normal-weight control subjects (both P < 0.001 for group effect; Fig. 3A and D, respectively). PCr-to-Pi and ATP-to-Pi ratios were likewise reduced in obese compared with normal-weight participants (PCr-to-Pi: P = 0.014, ATP-to-Pi: P = 0.003; Fig. 3B and E, respectively), whereas Pi content was similar in the two groups (P = 0.143 for group effect; Fig. 3C).

Comparison of baseline-adjusted neuroenergetic changes in obese subjects revealed no significant differences between tDCS and sham stimulation (PCr: P = 0.375, ATP: P = 0.113, Pi: P = 0.347, PCr-to-Pi ratio: P = 0.317, and ATP-to-Pi ratio: P = 0.213, respectively, for treatment by time interaction; Fig. 4A–E). In this group, tDCS showed a tendency to cause a drop in the PCr-to-Pi ratios 105 min after the end of the stimulation period relative to the sham condition (P = 0.088 for treatment by time interaction; Fig. 4B), whereas PCr, ATP, and Pi values as well as ATP-to-Pi ratios did not differ between the tDCS and the sham condition in obese participants (all P > 0.213).

Analyses in normal-weight volunteers revealed significantly different PCr and ATP levels as well as PCr-to-Pi and ATP-to-Pi ratios after tDCS versus sham stimulation (PCr: P = 0.023, ATP: P = 0.048, PCr-to-Pi: P < 0.001, and ATP-to-Pi: P < 0.001 for treatment by time interaction, respectively; Fig. 4A, B, D, and E), without any differences in Pi content (P = 0.101; Fig. 4C). Compared with sham stimulation, tDCS caused an early decline in PCr and PCr-to-Pi (P = 0.021 and P < 0.001 for treatment by time interaction), followed by an increase above the level of the sham condition (P = 0.010 and P < 0.001 for treatment by time interaction) and return to baseline levels at the end of the clamp experiment (P = 0.756 and P = 0.627 for treatment by time interaction, respectively; Fig. 4A and B). In addition, there was an initial fall in ATP values and ATP-to-Pi ratios after tDCS versus sham stimulation (P = 0.007 and P < 0.001 for treatment by time interaction), followed by a rapid rise (P = 0.004 and P < 0.001 for treatment by time interaction, respectively) and return to baseline values thereafter (P = 0.348 and P = 0.781 for treatment by time interaction; Fig. 4D and E).

Interindividual comparisons of tDCS-induced effects between the two participant groups revealed that, in contrast to normal-weight control subjects, obese men did not display the initial drop in PCr and ATP values or in PCr-to-Pi and ATP-to-Pi ratios, which occurred 65 min after the stimulation period ended in normal-weight men (PCr: P = 0.007, ATP: P = 0.004, PCr-to-Pi: P = 0.014, ATP-to-Pi: P = 0.002 for time-by-group interaction, respectively; Fig. 4A, B, D, and E). ATP levels and ATP-to-Pi ratios did not change in obese participants during the experiments (P = 0.711 and P = 0.899 for time main effect, respectively; Fig. 4D and E). However, we found a delayed and blunted drop in PCr values and PCr-to-Pi ratios 105 min after the stimulation period (P = 0.027 and P = 0.003 for time-by-group interaction; Fig. 4A and B), which was followed by a return to baseline levels by the end of the clamp (P = 0.995 and P = 0.885 for time-by-group interaction).

Correlation analyses between body weight–adapted glucose infusion rates and high-energy phosphate ratios involving both groups resulted in a significant negative relationship (r = −0.461, P = 0.010, and r = −0.421, P = 0.021, respectively, Fig. 5); that is, the higher the cerebral energy content, the lower the overall glucose infusion rates in both groups.

Obese participants showed smaller gray matter volumes in frontotemporal brain structures, anterior cingulum, putamen, insula, and cerebellum compared with normal-weight volunteers (Table 2 and Fig. 6A). Regression analyses of VBM measures with BMI revealed a strikingly similar pattern in these structural changes, that is, a negative relation between frontotemporal brain structures, anterior cingulum, putamen, and insula, as well as cerebellum and BMI, in both groups (Table 3 and Fig. 6B), confirming the previous observation of gray matter atrophy in obesity.

Multiple regression analyses between gray matter volumes and the cerebral high-energy phosphate content revealed a relationship between the neuroenergetic content and volumes in the right insula, putamen, and cerebellum (P = 0.007; Fig. 6C). Post hoc analyses in obese individuals, including gray matter volumes of the right insula and high-energy phosphates, revealed significant correlations in ATP (r = 0.39, P < 0.001), PCr (r = 0.27, P = 0.021), ATP-to-Pi ratio (r = 0.58, P = 0.022), and PCr-to-Pi ratio (r = 0.34, P = 0.007; Fig. 6C).

Our data demonstrate blunted neuroenergetic reactivity and diminished systemic glucose uptake in response to electrical brain stimulation in obese compared with normal-weight volunteers. Moreover, analyses revealed reduced gray matter volumes in the frontotemporal brain areas, anterior cingulum, putamen, insula, and cerebellum of obese men. In turn, insular gray matter volumes in obese participants correlated with the cerebral high-energy phosphate content, which suggests that decreased ATP and PCr levels may underlie the reduced appetite- and taste-processing area morphometry in these individuals.

This is the first study showing a reduced response of brain energy variability and systemic glucose uptake to tDCS in obesity. Although normal-weight participants display a biphasic course of high-energy phosphate levels with an initial drop, followed by a rise above baseline values upon tDCS, obese volunteers did not show any significant stimulation-induced variance in cerebral energy metabolism. It was only after 105 min (i.e., strikingly delayed) that we found a tendency for a drop in the PCr-to-Pi ratio in the obese individuals, which returned to baseline values by the end of the experiment. Because our observations preclude any conclusion in terms of underlying mechanisms, we can only speculate about the reason for this neuroenergetic rigidity in obese individuals. In view of the preexisting lowered ATP and PCr levels in obese compared with normal-weight individuals (7), the blunted neuroenergetic response to tDCS could be interpreted as a counterregulatory provision to prevent any further decline in the cerebral ATP content. In this regard, the reduced response to tDCS in obesity would represent a neuroprotective mechanism.

However, the neuroenergetic rigidity in obese individuals was accompanied by reduced gray matter volumes within frontotemporal brain structures, anterior cingulum, putamen, insula, and cerebellum compared with normal-weight control subjects, which is consistent with previous data providing evidence of obesity-related structural changes within the prefrontal cortex and the ventral striatum (19,20). Again, the question arises as to which mechanism might be behind the obesity-related brain morphology changes. At this point, it is conceivable that the observed morphometric alterations may relate to the reduced ATP and PCr levels in obesity. The volume of the right insula was correlated with the cerebral high-energy phosphate content; that is, the morphometric decline is apparently accompanied by a neuroenergetic undersupply, which is at least true for this specific region. One could therefore hypothesize that the disturbed energy homeostasis might have caused the morphometric shrinking in obesity. This seems likely because, strikingly, all of the volume-reduced brain areas are involved in food intake regulation via control of food reward (21), taste processing (22), and motivational goal-directed behavior (23), which is relevant in terms of food preferences. In this context, the right anterior dorsal insula and the dorsal midinsula fulfill an integrative function in the olfactogustatory system (24) and play an important role in flavor identification (25). The orbitofrontal cortex, in turn, is involved in the regulation of impulse control, eating behavior, and meal termination (21,22). Thus, it may be speculated that morphometric alterations in these brain areas due to a neuroenergetic decline are associated with abnormal sensory perception (26) and, in consequence, enhanced consumption of highly palatable foods (i.e., sweet or high-fat nutrients). In the long-term, the subsequent increase in calory intake would result in body weight gain and obesity.

This reasoning is consistent with previous data showing a negative relationship between gray matter volumes, particularly in frontotemporal brain areas and the insula, and BMI (27). However, we cannot draw definite conclusions about the cause-and-effect relationship between body weight gain and obesity-related brain atrophy. On one hand, one could speculate that high-calorie intake may induce a modification of the brain, which, in turn, potentiates calorie intake and further alters the brain in a self-reinforcing cycle. On the other hand, a part of the human population remains at a normal weight despite the spread of high-calorie food round the world, suggesting that the cause-and-effect relationship between obesity and brain atrophy is much more complex. For example, there is evidence that chronic stress enhances food intake. Increasing stress in daily life and chronic hypothalamic-pituitaryadrenal axis hyperactivity are linked to enhanced appetite, consumption of palatable high-calorie foods, and obesity (28). On the other hand, stress per se causes damage to brain structures (29).

In our study, however, obese participants showed not only suppressed neuroenergetic reactivity upon tDCS but also lower systemic glucose uptake. We have previously demonstrated that tDCS increases systemic glucose tolerance in normal-weight volunteers and that glucose infusion rates relate to respective changes in the high-energy phosphate content (9). Despite the obvious relationship between cerebral energy metabolism and systemic glucose uptake, how the two are linked is not entirely clear. One explanation is provided by the “energy on demand” mechanism, which proclaims that the brain, as the only organ that is able to supply itself with energy, can allocate glucose from the peripheral blood circulation to satisfy its own needs (30). In case of a neuroenergetic drop (e.g., as occurs upon tDCS), the brain would therefore request higher amounts of glucose from the systemic circulation, which becomes manifest in increased glucose infusion rates (31). In line with this, we found a negative correlation between ATP and PCr levels and glucose infusion rates during the hyperinsulinemic-euglycemic clamp in our study (i.e., the higher the brain energy levels, the lower the glucose uptake), which indicates that individuals with already high cerebral energy content need lower glucose infusion rates to meet their cerebral energy requirements. Given that this indeed is a physiological process, this fundamental mechanism seems to be disturbed in obesity and may be part of the observed neuroenergetic undersupply.

In conclusion, our data show an overall reduction in cerebral energy content with blunted reactivity of cerebral ATP and PCr levels and reduced systemic glucose uptake in response to tDCS in obese compared with normal-weight volunteers. As well as expanding our knowledge of lowered multifocal gray matter volumes in obesity, our study reveals that these changes are associated with reduced brain energy levels. Overall, our data suggest that a diminished brain energy supply may underlie the volume reduction of brain areas assigned to taste and appetite regulation and, therefore, to body weight gain. More generally, our study provides further support for the assumption that alterations in cerebral energy homeostasis may lead to enhanced food intake behavior and, in the long term, to obesity.

Acknowledgments. The authors thank Heidi Ruf and Martina Grohs, Department of Neuroendocrinology, University of Luebeck, Luebeck, Germany, for hormonal measurements.

Funding. This work was supported by funding from the German Research Foundation (DFG, KFO 126).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. K.J.-C., F.B., K.R., U.S., and K.M.O. interpreted the data and wrote the manuscript. K.J.-C., K.R., U.H.M., and K.M.O. analyzed the data. F.B. and K.M.O. conceived and designed the study. M.L., G.J., and U.H.M. collected the data. All authors edited and approved the final version for submission. K.J.-C. is the guarantor of this work and, as such, had full access to all data in the study and takes responsibility for the integrity and the accuracy of the data analysis.