Regional brain glucose metabolism is differentially affected by ketogenic diet: a human semiquantitative positron emission tomography

Ketogenic diet (KD) is recommended to avoid intense [18F]FDG myocardial physiologic uptake in PET imaging. Neuroprotective and anti-seizure effects of KD have been suggested, but their mechanisms remain to be elucidated. This [18F]FDG PET study aims to evaluate the effect of KD on glucose brain metabolism. Subjects who underwent KD prior to whole-body and brain [18F]FDG PET between January 2019 and December 2020 in our department for suspected endocarditis were retrospectively included. Myocardial glucose suppression (MGS) on whole-body PET was analyzed. Patients with brain abnormalities were excluded. Thirty-four subjects with MGS (mean age: 61.8 ± 17.2 years) were included in the KD population, and 14 subjects without MGS were considered for a partial KD group (mean age: 62.3 ± 15.1 years). Brain SUVmax was first compared between these two KD groups to determine possible global uptake difference. Semiquantitative voxel-based intergroup analyses were secondarily performed to determine possible inter-regional differences by comparing KD groups with and without MGS, separately, to 27 healthy subjects fasting for at least 6 h (mean age of 62.4 ± 10.9 years), and KD groups between them (p-voxel < 0.001, and p-cluster < 0.05, FWE-corrected). A 20% lower brain SUVmax was found in subjects under KD with MGS in comparison to those without MGS (Student’s t-test, p = 0.02). Whole-brain voxel-based intergroup analysis revealed that patients under KD with and without MGS had relative hypermetabolism of limbic regions including medial temporal cortices and cerebellum lobes and relative hypometabolism of bilateral posterior regions (occipital), without significant difference between them. KD globally reduces brain glucose metabolism but with regional differences, requiring special attention to clinical interpretation. On a pathophysiological perspective, these findings could help understand underlying neurological effects of KD through possible decrease of oxidative stress in posterior regions and functional compensation in the limbic regions.


Introduction
The brain is a major consumer of body energy, accounting for at least 20% of the whole energy consumption, while it is only 2% of body weight [1,2]. Most of this energy is produced by aerobic metabolism utilizing glucose, with fat being a very low energy source under regular conditions. Brain glucose metabolism can be affected by many parameters in physiological and pathological conditions, for example, age, glycemia, local damage, and neurodegenerative disorders such as Alzheimer's disease (AD). Ketogenic diet (KD) certainly also influences brain glucose metabolism.
KD consists of reducing carbohydrate intake in favor of high-fat consumption to induce glucose metabolism This article is part of the Topical Collection on Neurology.

3
shifting toward fatty acid metabolism. This last metabolism uses ketone bodies (mainly β hydroxybutyrate and acetoacetate) as a source of energy instead of glucose [3,4]. KD has been proven beneficial in positron emission tomography (PET) imaging to avoid intense FDG myocardial physiologic uptake [5,6]. For this reason, KD is done prior PET examinations indicated for endocarditis, sarcoidosis, or any disorder for which myocardial glucose suppression (MGS) is necessary. Interest in KD as a therapeutic strategy has risen since it was first suggested useful in pharmacoresistant epilepsy in the 1920s [7,8]. KD could have potential utility as an add-on therapy in several neurologic disorders, such as gliomas, head trauma, sleep disorders, stroke, and neurodegenerative disorders (AD, Parkinson's disease, multiple sclerosis) [4,[9][10][11][12][13]. However, the underlying mechanisms of the antiepileptic and neuroprotective effects of KD are still not fully understood [14].
On a cellular level, there is evidence that KD reduces reactive oxygen species and increases ATP and mitochondrial biogenesis [3,15]. KD also leads to an increase in polyunsaturated fatty acids that regulate neuronal membrane excitability by blocking voltage-gated sodium channels [16] and modify glycolysis.
Preclinical and clinical studies have investigated the effects of ketone bodies on the brain and the interplay between glucose and ketone body metabolism. It is indeed assumed that during KD, ketone bodies are the main source of energy in the brain. This has been shown by catheterization of cerebral vessels in fasting humans [17], a reduction in cerebral metabolic rates of glucose (CMRglc) in ketotic rats on PET in the cortex and cerebellum using the Gjedde-Patlak model when plasma ketone bodies increased [18], and a proportionate decrease in CMRglc in humans with blood ketone increases [19]. Even if most authors agree [4,[18][19][20] that a global decrease in cerebral glucose metabolism occurs under KD toward the use of ketone bodies [21], the hypothesis that different brain regions have distinct utilization of glucose is still poorly investigated. Some authors found an increased cerebral metabolic rate of acetoacetate and a decreased cerebral metabolic rate of glucose in all regions [20], which was more marked in the caudal middle frontal cortex, while others found a preserved uptake in the basal ganglion and cerebellar hemispheres but did not identify regionally reduced uptake areas [22]. A recent PET study from Bennett and colleagues [23] focused on correlations between blood glucose and ketones levels and brain glucose metabolism after a short KD. They calculated SUVmean values normalized for bodyweight (SUVbw) in four regions of interest (whole brain, precuneus, cerebellum, and basal ganglia). They found a negative correlation between SUVbw in the whole brain and glucose and ketone serum levels, more pronounced in the precuneus, concluding that these regional differences should be taken into consideration for interpretation or quantification analyses [23]. Further intergroup studies should determine the possible impact of KD in terms of hypo/hypermetabolism. Precisely, the aim of our [ 18 F]FDG PET intergroup study is to evaluate regional brain glucose metabolic relative changes after a short KD compared to a control population, using a whole-brain semiquantitative voxel-based analysis.

Population studied
Patients who underwent a KD prior to brain and whole-body [ 18 F]FDG PET for suspected endocarditis between January 2019 and December 2020 in the Department of Nuclear Medicine at Timone Hospital in Marseille were retrospectively included. Patients were hospitalized in cardiology units, with medical and paramedical teams trained to KD rules. Our Nuclear Medicine department sends KD instructions to the units before examination. The dietary instructions were as follows: 12 h minimum fasting, ideally 15 to 18 h fasting before examination; no blood perfusion with glucose or parental nutrition 15 to 18 h before examination; and a KD performed for 36 to 48 h before examination. The minimum last two meals before fasting should be high-fat and low-carbohydrate meals. A list of permitted and banned foods is provided with menu examples.
Exclusion criteria were the following: presence of cerebral abnormalities on PET, CT, or MRI, such as brain embolies or abscess; brain sequelae (stroke, surgery…); presence of neurologic symptoms suggestive of encephalitis or neurodegenerative disorders. Apparent normal brain PET on visual interpretation performed by two different independent readers was selected. Among the 152 patients selected (Fig. 1), 103 patients were excluded because of brain abnormalities. One patient was excluded for being under 18 years old.
Myocardial metabolism was examined on whole-body PET for the 48 patients without brain abnormalities and used as an indirect marker of full KD adherence. Images were classified in terms of the presence or absence of significant visual myocardial uptake. In this way, 34 subjects were considered as the KD population with MGS, 38% female (mean ± SD age of 61.8 ± 17.2 years), and 14 patients were considered as the partial KD group in the absence of significant MGS (mean age: 62.3 ± 15.1 years). Brain and myocardial maximum standard uptake values (SUVmax) were collected.
A control group of 27 healthy subjects was also included. They had a minimum 6-h fasting, with a similar age and sex ratio as those of KD population (mean ± SD age of 62.4 ± 10.9 years; 40% of females). No significant age or sex differences were found using Student's t-test between the groups KD with MGS and healthy controls (p = 0.85 for age, and p = 0.87 for sex), between KD without MGS and healthy controls (p = 0.81 for age, and p = 0.34 for sex), and between KD with and without MGS (p = 0.92 for age, p = 0.25 for sex). They had no neurological or psychiatric disorder and a normal neuropsychological testing and brain MRI (clinical trial reference NCT00484523). Glycemia levels of the healthy subjects were individually below a threshold of 1 g/l. The study was approved by APHM institution with the reference PADS21-311.

[ 18 F]FDG PET acquisitions and reconstructions
A dose of 3 MBq per kilo of [ 18 F]FDG was injected followed by a 50-min rest in a dark quiet room for the KD population and a dose of 111 MBq followed by a 30-min rest for the healthy subjects. Brain PET and then whole-body PET for the KD population were performed by a GE Discovery PET/CT camera (GE Discovery 710 camera for both KD groups, and GE Discovery ST for the healthy subjects). Acquisition and reconstruction protocols were harmonized for comparability with a same matrix size and a same reconstructed resolution. Brain PET-CT was acquired using a three-dimensional acquisition of 7 min when associated with a whole-body PET acquisition (endocarditis group) and 15 min when the brain was acquired alone (healthy subjects), considering the difference in radiopharmaceutical activities, with an axial resolution of 6.2 mm and 47 contiguous transverse sections of the brain of 3.27-mm thickness. Images were reconstructed using an OSEM-type iterative reconstruction algorithm with 5 iterations and 32 subsets and corrected for attenuation using a scan.

Statistical analysis
Brain cortical SUVmax of the two KD groups were first compared using Student's t test to determine possible global uptake difference (Table 1). Secondly, semiquantitative voxel-based intergroup analyses were performed to determine possible inter-regional differences by comparing KD groups with and without MGS, separately, to the 27 healthy  subjects fasting for at least 6 h, and thereafter KD groups between them. Concerning semiquantitative voxel-based intergroup statistical analysis, images were processed using SPM8 (Welcome Department of Cognitive Neurology, University College, London, UK) run on MATLAB. Spatial normalization (MNI atlas) and smoothing with an 8-mm FWHM Gaussian filter were performed. A proportional scaling method was chosen to account for intersubject variability relying on different [ 18 F]FDG doses (111 MBq for brain alone or 3 MBq per kilo for brain and whole-body) and different times of acquisition after injection (30 min postinjection for 15 min versus 50 min postinjection for 7 min) [24]. Age and sex were added as covariables. A voxel p value of 0.001 was chosen for the voxel and combined with a p-cluster < 0.05, FWE-corrected (family-wise error). Glycemia before PET examinations was recorded and correlated with brain SUVmax using the Pearson correlation coefficient.

Semiquantitative voxel-based brain PET metabolic analyses
To analyze possible brain regional metabolic changes induced by a short KD, the following semiquantitative voxel-based intergroup analyses were performed: (i) KD with MGS to healthy subjects; (ii) KD without MGS to healthy subjects; (iii) KD with MGS to KD without MGS. All results are expressed with a p value < 0.001 for the voxel and a p-cluster < 0.05, FWE-corrected.
In comparison to healthy subjects (n = 27), KD with MGS (n = 34) was associated with relatively increased FDG uptake in the limbic regions, bilateral cerebellum lobes, and left motor cortex (Figs. 2 and 3; Table 2): right inferior temporal cortex (peak T-score = 7.01; k = 1307 voxels), left temporal inferior and medial cortex (peak T-score = 8.49; k = 3380 voxels), left middle frontal cortex (peak T-score = 7.03 and k = 1373), left medial frontal cortex (peak T-score = 6.07 and k = 1810), left inferior frontal and precentral cortex or motor cortex (peak T-score = 5.23; k = 663), in the left cingulate cortex (peak T-score = 6.94; Fig. 2 Whole-brain anatomical localization of significant metabolic changes after KD with MGS (n = 34) compared to healthy subjects (n = 27) in axial slices, with T-score scale on the left side. In orange, regions with increased metabolism include bilateral temporal cortex, left medial temporal cortex, left frontal cortex (middle, medial, inferior), left precentral cortex, left cingulate, and the bilateral cerebellum. Blue indicates regions with decreased metabolism, including in bilateral posterior regions (bilateral cuneus and right middle temporal) k = 1500), and in both cerebellum lobes: left cerebellum (peak T-score = 5.45; k = 852) and right cerebellum (peak T-score = 5.51; k = 695). KD with MGS was also associated with a relatively decreased FDG uptake in bilateral posterior regions (bilateral cuneus and right middle temporal; peak T-score = 5.65; k = 6099).

Discussion
This study shows regional brain glucose metabolic changes under a short KD prior to PET examination. Using proportional scaling, and in comparison to a healthy group, we found under KD a relatively increased FDG uptake mainly in bilateral limbic regions and cerebellum cortices, and a relatively decreased FDG uptake in posterior regions (bilateral cuneus and right middle temporal). To date, only a few studies have investigated brain glucose metabolic changes during KD [20][21][22][23], with different approaches for activity normalization and very distinct number of inclusions and KD durations, including correlation and intergroup study in comparison to a control group. In a recent correlation study, Bennet et al. [23] investigated 52 patients who underwent a short KD (48 h with 18-h fasting) prior to a PET for possible cardiac sarcoidosis or suspected intracardiac infection. They calculated SUVmean values normalized for bodyweight (SUVbw) for four selected regions of interest: the whole brain, precuneus, basal ganglion, and cerebellum. The study focused on biological and PET imaging correlations, without intergroup analysis. Interestingly, their adjusted linear regression analysis found a negative correlation between SUVbw in the whole brain and glucose and ketone serum levels, even more marked in the precuneus. These correlation results could be articulated with the results from our intergroup whole-brain voxel-based study that found a relative hypometabolism in the occipital regions under KD.
On the other hand, using visual interpretation and quantification analysis, Korsholm et al. [22] found a global hypometabolism of the cortex with preserved uptake in the basal ganglia and the cerebellum in one 11-year-old child with drug-resistant epilepsy on a two-and-half-month KD. No areas with focally reduced uptake were identified, but there was only one subject, and the effects could be partially explained by seizure relief itself (patient with epilepsy since the age of 5, last seizure day prior PET). Also, long-term KD might modify brain metabolism in a different manner than a short KD. The transcription regulation of some transporters or enzymes that control cell glucose and ketone uptakes might need a long-term KD, as for example, downregulation of GLUT transporters [4]. Finally, children are known to have different brain metabolism than adults. Courchesne-Loyer et al. [20] found on a sample of 10 young healthy adults (aged 35 ± 15) that a 4-day KD decreased CMRglc in all regions with the greatest decrease in the caudal middle frontal cortex. They compared the same subjects pre-to post-KD CMRglc using the quantification method of the multiple time graphical analysis of Patlak et al. This method is more sensitive to global modifications than regional ones as proportional scaling. We found a significant 20% lower global brain SUVmax in the KD population without MGS compared to the KD population with MGS. These results are in concordance with those found in the literature, since most authors agree that there is a lower brain glucose uptake under KD [4,[18][19][20]23]. Nevertheless, we found out in the semiquantitative voxel-base analysis that the KD group without MGS had a similar metabolic pattern than the KD group without MGS, showing that even a partial KD could lead to brain regional modifications. These results indirectly confirm that the KD population without MGS can be considered as a "partial KD population." These results could also illustrate that the brain and myocardium do not carry the same mechanisms for glucose intake, probably relying on the fact that brain's glucose uptake is mostly insulin-insensitive. In a complementary way, Bennett et al. found a relation between brain regional glucose hypometabolism and blood glucose and ketone serum levels.
On a cellular level, regional metabolic changes under KD probably rely on two opposite mechanisms that regulate Fig. 4 Whole-brain anatomical localization of significant metabolic changes after KD without MGS compared to healthy subjects in axial slices, with T-score scale on the left side. In orange, regions with increased metabolism include left temporal cortex, left medial frontal cortex, left cingulate. Blue indicates regions with decreased metabolism, including in bilateral occipital regions Table 3 Results of the semiquantitative analysis comparing the KD group without MGS to healthy subjects (p value < 0.001 uncorrected for the voxel, and a p-cluster < 0.05 FWE-corrected). X, y, z correspond to Talairach coordinates (mm). Unc uncorrected, Corr corrected No significant brain regional metabolic changes were found by comparing KD groups with and without MGS glucose cell intake [4]. The first is the decrease in glucose cell uptake by the glucose transporter GLUT due to the accumulation of glycolytic intermediates secondary to an increase in ketone bodies. The second mechanism is the increase in cerebral blood flow that probably increases glucose uptake [4]. Even if the predominant mechanism under KD is a reduction in glucose uptake in favor of ketone bodies, a possible increase in cerebral blood flow in some regions could account for the regional differences observed under KD. Increased cerebral blood flow has been reported as a functional compensation phenomenon in case of decrease metabolism, for example in AD [25]. The downregulation of GLUT expression under KD is probably observed under long-term KD [4].
In addition, different cell types in the brain utilize glucose differently. In cases of stress, or high glucose demand, astrocytes can use glycogen to spare glucose for neurons [25]. From a physiopathology point of view, neuroprotective and anti-seizure effects of KD probably have also different mechanisms (on molecular, cellular, and regional levels). The pathophysiology of these diseases is different, and a single effect of KD probably cannot account for all its potential effect. KD effects in neurodegenerative disorders are linked to the global decrease in brain FDG uptake, implying less oxidative stress. Interestingly, we found an even more marked hypometabolism in the posterior cortex under KD, regions that are affected in neurodegenerative disorders such as AD. A relative decreased metabolism predominant in the posterior regions under KD might be an indirect sign that these regions are less exposed to oxidative stress, and not necessarily show that these regions are further impaired as in the evolution of AD. On the other hand, the relative increase in glucose uptake in limbic regions, including temporal medial regions, might reflect functional compensation [26,27]. Functional compensation in the limbic regions is a well-descripted process [26,28], and KD might help this process. Further studies with specific designs needed to confirm these hypotheses.
In addition, during normal aging, there is a global CMRglc decrease in the brain gray matter, and even more in AD while brain metabolic rate of acetoacetate is stable (CMRac) [29]. A supplementary evaluation of CMRac, in parallel to that of glucose, could help understand further KD pathophysiology.
The present study has several limitations mainly due to its retrospective design. The fact that all patients who underwent a KD had suspicion of endocarditis could have been a recruitment bias since neurologic complications of endocarditis are found in approximately 25% of patients [30]. However, we only selected patients without neurologic symptoms or CT/MRI abnormalities on neuroimaging and with apparent normal brain PET on visual interpretation. Additionally, out of 34 patients, only 59% had confirmed endocarditis on follow-up data. It would be interesting to verify our results in a larger population.
Our semiquantitative approach is justified by the objective to determine possible inter-regional differences (does KD differentially affect brain regional metabolism, or with a similar global decreased effect?). We therefore performed a proportional scaling, specifically focusing on possible regional metabolic changes. Even if this technique has some limitations, such as underestimating the extent of hypometabolic areas [24], it is also justified considering that our two groups had different FDG doses and different times of acquisition after injection (i.e., the relative comparison of regional metabolism between 2 groups of patients with distinct whole-brain metabolism).
Additionally, the KD duration was 36-48 h before examination, and it has been shown that a longer KD duration has a stronger effect on brain metabolism in rodents [19] and a stronger effect on MGS in humans [6]. It would be valuable to perform a clinical study with different KD duration, to estimate the necessary KD duration to induce significant brain regional metabolic changes, and if a long-term KD differentially affects brain metabolism. Also, a prospective study with detailed questionnaire of ingested meals or proposed meals would be a more accurate way to evaluate different KD than MGS, and in relationship to glycemia.

Conclusion
KD globally reduces brain glucose metabolism but with regional differences. It causes a relative hypermetabolism of limbic regions, including medial temporal cortices, and a relative hypometabolism of occipital regions. These results have a clinical implication, since the realization of a KD could lead to misinterpretation of these regions, and the findings also provide a pathophysiological perspective to help us understand underlying neurological effects of KD through possible on neuroprotection and functional compensation.
Funding The local PET database of healthy controls was funded by APHM (NCT00484523).

Data availability
The PET data that support the findings are available from the corresponding author upon reasonable request. The datasets generated during analysis are available from the corresponding author on reasonable request. Please note that a pre-print version of the manuscript is available at Research Square [https:// doi. org/ 10. 21203/ rs.3. rs-19050 69/ v1]

Declarations
Ethics approval and consent to participate The retrospective observations required no ethical approval requirement other than informed consent. The local PET database of healthy controls was acquired in accordance with the Declaration of Helsinki, with informed written consent from the patients and approval from the "CPP Sud