Differential utilization of ketone bodies by neurons and glioma cell lines: a rationale for ketogenic diet as experimental glioma therapy

High-grade gliomas are intrinsic brain tumors characterized by resistance to apoptotic stimuli, diffuse infiltration into the surrounding tissue and local immunosuppression. Despite advances in research on tumor biology and efforts to promote new therapies, the prognosis for patients with high-grade gliomas is still poor. Currently available treatment options for glioblastoma patients, including surgery, radio- and chemotherapy, result in a median survival of only about 12 months [1, 2]. Obviously, other therapeutic approaches are needed that, on the one hand, impair tumor cell growth and, on the other hand, permit an adequate quality of life.

Many malignant cells display high rates of glycolysis and lactate production, even in the presence of adequate oxygen, a phenomenon known as aerobic glycolysis or the Warburg effect [3]. Additionally, tumor hypoxia results in constitutive upregulation of glycolysis and is considered to substantially contribute to the resistance of tumor cells to therapeutic strategies [4, 5]. One possibility to affect the metabolism of such "glucose dependent" tumors could be the ketogenic diet. The classic ketogenic diet is a high-fat and low-carbohydrate dietetic approach raising levels of serum ketone bodies, i.e. acetoacetate, 3-hydroxybutyrate (from less than 0.1 mM to 0.2-1.8 mM and 2-5 mM, respectively) and acetone, and lowering brain glucose uptake [6‐8]. Acetoacetate and 3-hydroxybutyrate are almost exclusively synthesized in the liver from acetyl-CoA that results from the beta-oxidation of fatty acids. Acetone is formed from acetoacetate by spontaneous decarboxylation and is generally considered of little metabolic significance. Energy generation from ketone bodies takes place via the citric acid cycle and oxidative phosphorylation and therefore requires proper mitochondrial function. In contrast to glucose, ketone bodies thus bypass cytoplasmic glycolysis and directly enter the citric acid cycle as acetyl-CoA. In fasting humans, the water-soluble ketone bodies can supply approaching 60% of the brain's energy requirement. Besides being a source of energy, ketone bodies can provide substrates for anabolism, particularly for the synthesis of lipids such as cholesterol in myelin. A simplified diagram of ketone body metabolism is shown in Figure 1A. As malignant cells are thought to depend on glucose as a major source of energy while having impaired mitochondrial function [9‐11], a ketogenic diet thus might induce a tumor-selective energy deprivation. There has been considerable research on the safety, the anticonvulsant and neuroprotective effects of the ketogenic diet [8, 12, 13], whereas less attention has been paid to its potential for tumor therapy. Similar to tumors of peripheral tissues [14], glial tumors were found to have less activity of 3-oxoacid-CoA transferase 1 (OXCT1), the rate-limiting enzyme of ketone body degradation, than normal human white (and gray) matter [15]. Following encouraging observations in two female pediatric patients with advanced stage malignant astrocytoma who seemed to benefit from a ketogenic diet [16], a reduction of glioma growth was noticed in orthotopic mouse models when a calorie-restricted ketogenic diet was applied [17, 18]. A first clinical study assessing the feasibility and safety of a non-calorie-restricted ketogenic diet in patients with recurrent glioblastoma (the ERGO trial, NCT00575146) revealed good tolerability and suggested some antitumor activity [19]. Nevertheless, to date, little data concerning a potential application of a ketogenic diet in brain tumor treatment is available. So far no randomized controlled trials have been initiated on the ketogenic diet for tumor therapy. Further, biological effects and associated metabolic alterations of a ketogenic diet on glioma versus neural cells remain to be elucidated. In the present study, we therefore characterized in vitro effects primarily of 3-hydroxybutyrate, the major circulating ketone body, on five human glioma cell lines (U87MG, U251MG, LNT-229, T98G and A172) and rat hippocampal neurons and examined the efficacy of a non-calorie-restricted ketogenic diet in an orthotopic glioma xenograft mouse model.

Ketone body metabolizing enzymes are expressed in the five glioma cell lines both at the mRNA and the protein level. First, we examined whether the five glioma cell lines express key enzymes involved in ketone body metabolism. U87MG, U251MG, LNT-229, T98G and A172 cells exhibited expression of 3-hydroxybutyrate dehydrogenase (BDH), 3-oxoacid-CoA transferase (OXCT1), acetyl-CoA acetyltransferase (ACAT1) and acetoacetyl-CoA synthetase (AACS) at the mRNA (Figure 1B) and the protein level (Figure 3B). No consistent change in expression was observed following exposure to 3-hydroxybutyrate for 24 h or 48 h (Figure 3B and data not shown). In rat hippocampal neurons, the expression of these enzymes was confirmed by real-time quantitative PCR (data not shown).

3-hydroxybutyrate protects primary rat hippocampal neurons but not human glioma cell lines against glucose deprivation-induced cell death. Having confirmed the expression of ketone body metabolizing enzymes, we assessed whether 3-hydroxybutyrate could serve as an alternative energy source under conditions of reduced glucose availability. For this purpose, rat hippocampal neurons or glioma cells were incubated in the absence or presence of 5 mM 3-hydroxybutyrate in serum-free medium containing different glucose concentrations (hippocampal neurons: 0 mM, 1 mM, 2 mM or 5 mM; glioma cell lines: 0 mM, 1 mM, 2.5 mM, 5 mM, 10 mM or 25 mM), and cell density was analyzed by MTT assay (Figure 2A-B) or crystal violet staining (Figure 2C). Primary rat hippocampal neurons benefited from the presence of 3-hydroxybutyrate by prolonged survival under glucose deprivation. By contrast, in no glioma cell line, a rescue from glucose withdrawal-induced cell death by this ketone body was detectable. In untransformed primary rat astrocytes, some advantage of 3-hydroxybutyrate supplementation was evident, although to a lesser extent than in primary rat hippocampal neurons (Figure 2D).

Ketone bodies do not alter growth or proliferation of glioma cell lines. Next, we analyzed potential effects of ketone bodies on cell growth and proliferation in the presence of sufficient glucose. In all glioma cell lines tested, the presence of ketone bodies, i.e. acetoacetate, 3-hydroxybutyrate, or both, did not alter cell density at different time points up to 84 h after exposure (Table 2 and data not shown). Furthermore, cell proliferation as assessed by BrdU-uptake was unaffected by 3-hydroxybutyrate (data not shown). As glucose availability has been demonstrated to be relevant for cellular survival under hypoxic conditions [35‐37], we investigated whether hypoxia might alter the deficiency of glioma cells to metabolize ketone bodies. However, at different oxygen (21%, 1% or 0.1% O₂) or glucose concentrations, offering 3-hydroxybutyrate did not influence cell density (Table 2 and data not shown). Furthermore, the presence of 3-hydroxybutyrate did not modify glucose consumption or lactate generation of the glioma cell lines examined (analyzed up to six days after exposure; Additional file 1, Figure S1). Because 3-hydroxybutyrate might be metabolized for cellular processes which would not directly be reflected by differences in glucose consumption, 3-hydroxybutyrate levels were measured in cell culture supernatants. Only a marginal decrease in 3-hydroxybutyrate in the glioma cell culture supernatants was observed (between 0.0 mM and 0.5 mM; analyzed for up to six days after exposure), indicating no substantial utilization of this additional energy source. In cultures of rat hippocampal neurons, 3-hydroxybutyrate supernatant concentrations were reduced to a greater extent despite lower cell densities (1.3 mM decrease after three days of exposure). A loss of p53 function has been shown to result in metabolic alterations such as reduced cellular respiration, increased glucose consumption and lactate production [38]. Therefore, we examined whether antagonizing p53 would modulate the metabolism of 3-hydroxybutyrate. However, cell density of both p53 wild-type LNT-229 cells and of LNT-229 cells depleted of p53 by stable shRNA expression (LNT-229 p53sh) did not differ in the absence or presence of 3-hydroxybutyrate (Table 2 and data not shown), indicating no induction of ketone body metabolism by inhibition of p53 function.

Exposure to 3-hydroxybutyrate does not modulate glioma cell motility and invasiveness. The infiltrative behavior of malignant glioma cells is a function of two phenotypes, migration and invasiveness. Migration refers to the capacity of locomotion whereas invasion involves migration plus a degradative function achieved by the liberation of proteolytic enzymes. Using classical migration and matrigel invasion assays, we did not detect any significant changes when 3-hydroxybutyrate was offered (Table 3).

Treatment with 3-hydroxybutyrate does not modify the expression of hypoxia-inducible factor-1α (HIF-1α) and of its target genes. Human and murine solid tumors are characterized by hypoxic areas with intratumoral pO₂ values between 2 and 12 mmHg (approximately 0.3% - 2% O₂; [39‐41]). We therefore looked at the expression of ketone body metabolizing enzymes under hypoxia (1% or 0.1% O₂) as well. Regardless of the absence or presence of 3-hydroxybutyrate, we noticed a considerable downregulation of BDH, OXCT1 and ACAT1 mRNA levels in all cell lines (Figure 3A and data not shown) in hypoxia. At 24 h after 3-hydroxybutyrate exposure, this downregulation was visible on protein level primarily in U251MG cells (Figure 3B). Succinate is generated during the activation of acetoacetate to acetoacetyl-CoA by the enzyme OXCT1. Accumulation of succinate can result in the stabilization of HIF-1α via product inhibition of prolyl hydroxylase (PHD) enzymes [42]. The small amounts of 3-hydroxybutyrate which might be metabolized by glioma cells (between 0.0 mM and 0.5 mM decrease in supernatant level; see above) could modulate HIF-1α expression as an indirect hint for ketone body degradation. However, in these cell lines, 3-hydroxybutyrate did not induce an elevation of HIF-1α protein (Figure 3B). Similarly, we did not observe a rise in HRE reporter gene activity (Figure 3C) or in expression of the HIF-1α target genes GLUT1, VEGF and MCT4 (Additional file 2, Figure S2, and data not shown) in any of the five glioma cell lines examined. 3-nitropropionic acid (3NPA) is an irreversible inhibitor of succinate dehydrogenase, leading to elevated succinate levels. As shown in Figure 3D, 3NPA led to an increase in HIF-1α protein, at least under hypoxic conditions, confirming that HIF-1α is indeed regulated by succinate. In summary, these results suggest that 3-hydroxybutyrate does not modulate HIF-1α expression or activity in the glioma cell lines analyzed, and therefore provide additional evidence for a defective ketone body utilization of glioma cells.

3-hydroxybutyrate does not influence glioma cell sensitivity towards different triggers of cell death. Finally, we were interested in whether ketosis could change the susceptibility to different mechanisms of cell death. In U87MG, U251MG, LNT-229, T98G and A172 cells, sensitivity to treatment with rotenone, an inhibitor of mitochondrial respiratory chain complex I, or 3-bromopyruvate, a synthetic brominated derivative of pyruvic acid inhibiting the enzyme hexokinase II, was not altered in the presence of 3-hydroxybutyrate (Figure 4A-B and data not shown). Moreover, 3-hydroxybutyrate did not modify cell death induction by activation of the mitochondrial apoptotic pathway via tumor necrosis factor-related apoptosis-inducing ligand (TRAIL; Figure 4C and data not shown), or inhibition of clonogenic cell growth by temozolomide, the adjuvant standard of care chemotherapeutic agent in glioma patient treatment (Figure 4D and data not shown). In all five glioma cell lines examined, neither synergistic nor antagonistic effects of 3-hydroxybutyrate were detected.

The ketogenic diet is well accepted and induces stable ketosis in vivo, but does not affect tumor growth or survival. In a translational approach, the present study aimed to assess some characteristics of an unrestricted ketogenic diet similar to the one offered to patients in the pilot study carried out in our department (NCT00575146). Using the same ketogenic diet, Otto et al. had observed a delayed growth of subcutaneously implanted tumors of the gastric adenocarcinoma cell line 23132/87 [31]. After ruling out major tumor-supporting effects of ketone bodies in vitro, we performed an orthotopic xenograft mouse model experiment to assess whether this ketogenic diet might result in a tumor-selective energy deficit and thus reduce tumor growth in vivo. After orthotopic inoculation of LNT-229 glioma cells, mice were randomized into two non-calorie-restricted diet groups, a standard diet group and a ketogenic diet group. All animals readily accepted the unrestricted ketogenic diet. Similar to the findings of other studies [18, 31], their body mass did not differ substantially from those of control animals (Figure 5A). Blood 3-hydroxybutyrate levels were continuously and significantly higher in mice fed the ketogenic diet than in animals on the control diet. By contrast, no relevant difference in blood glucose values was observed between the two groups (Figure 5B). Interindividual and intervisit blood glucose levels varied strongly. Similarly, serum concentrations of insulin-like growth factor 1 (IGF-1) were not significantly different between the two diet groups (Table 4). MRI analysis and observation for symptom-free survival did not reveal a difference between the two diet groups regarding tumor size and survival (Figure 6A). Histological analysis did not show differences in the proliferation rate (Figure 6B), nor in size or morphology of the tumors (data not shown). ATP, glucose or lactate concentrations in normal and tumor tissues did not differ significantly between the two diet groups (metabolite concentrations in tumor areas, mean and standard deviation, μmol/g, ATP: control, 1.9 ± 0.9, ketogenic diet, 1.9 ± 0.6, glucose: control, 1.6 ± 0.8, ketogenic diet, 2.0 ± 0.8, lactate: control, 19.8 ± 5.1, ketogenic diet, 20.9 ± 3.2; Figure 7). Consistent with the findings of other studies [43, 44] but not reaching statistical significance, bioluminescence imaging data suggested lower glucose and higher ATP and lactate concentrations in tumors than in corresponding normal tissues (tumor-to-normal tissue ratio, mean and standard deviation, ATP: 1.59 ± 0.91, glucose: 0.88 ± 0.12, lactate: 1.29 ± 0.55).

Strategies specifically targeting the altered metabolic pathways of tumors are increasingly receiving attention [45, 46]. In contrast to new chemotherapeutic approaches, altering diet is perceived by many as an easy option to delay tumor progression with few side effects. To the best of our knowledge, however, no randomized clinical trials have been initiated on ketogenic diets for therapy in any tumor type. Concerning nutrition, patients see themselves faced with incomplete information and contradictory, sometimes extreme recommendations. Given the principle of primum nil nocere, we therefore considered it reasonable and necessary to preclinically evaluate effects of dietary changes just like those of other anticancer drugs.

Cerebral ketone body metabolism depends on (i) concentrations in blood, (ii) the transport across the blood-brain barrier and into cells and (iii) the activity of ketone body metabolizing enzymes [6]. Blood concentration is considered the most important factor affecting the rate of cerebral ketone body metabolism. As the blood-brain barrier is relatively impermeable to most hydrophilic substances, transporters for short-chain monocarboxylic acids (MCTs, SLC16 family [47]), such as ketone bodies and lactate, are needed and govern access of acetoacetate and 3-hydroxybutyrate to central nervous system tissues [48]. Ketone body entry into brain cells occurs by diffusion (at a moderately high rate) and carrier-mediated processes (MCTs; probably less regulated). Finally, ketone body metabolism depends on the activities of the relevant enzymes, but due to the limited number of studies [49], sufficient information concerning those enzymes in humans is not available.

We here demonstrate a deficient utilization of the major ketone body 3-hydroxybutyrate by human malignant glioma cells. 3-hydroxybutyrate did not alter cell density or proliferation and could not protect the human glioma cell lines examined against glucose deprivation. By contrast, this ketone body conferred protection from glucose withdrawal-induced cell death in primary rat hippocampal neurons (Figure 2). A recent study using organotypic rat hippocampal slice cultures supports this observation [50], being in accordance with the capability of rats to metabolize ketone bodies [51, 52]. Neuroprotective effects of 3-hydroxybutyrate on primary spinal cord neurons from SODI-G93A mice and on SH-SY5Y human dopaminergic neuroblastoma cells exposed to the complex I inhibitor rotenone have also been described [53, 54]. In our experiments, no protective effect of 3-hydroxybutyrate on glioma cell death induced either by rotenone or by the glycolysis inhibitor 3-bromopyruvate was detectable (Figure 4A-B). Hence, in those cell lines, this ketone body could not compensate for energy depletion induced by disturbance of the mitochondrial respiratory chain or glycolysis. Further, we did not observe proapoptotic effects of ketone bodies in glucose-free medium as did Skinner et al. using the human neuroblastoma cell line SK-N-AS [55], indicating that this effect might be cell type specific. An increase in HIF-1α protein levels caused by accumulation of succinate and inhibition of PHD enzymes (see Figure 1A) has been described in diet-induced ketotic as well as in 3-hydroxybutyrate-infused rat brain [56] and would probably be a rather unwanted effect regarding tumor treatment [57, 58]. We therefore evaluated expression, transcriptional activity and target gene modulation of HIF-1α. However, neither was affected by 3-hydroxybutyrate (Figure 3 B-C; Additional file 2, Figure S2), suggesting that this ketone body is not metabolized in the glioma cell lines examined. Furthermore, 3-hydroxybutyrate did not influence the migratory behavior of the glioma cell lines (Table 3). By contrast, it has been observed that 3-hydroxybutyrate like lactate may function as a chemoattractant, stimulating the migration of MDA-MB-231 human breast cancer cells in vitro [59].

Together, our in vitro experiments did not indicate any ketone body metabolizing activity of the glioma cell lines examined, a finding which contrasts the protective effect of 3-hydroxybutyrate on rat hippocampal neurons. The observed phenotype is probably not caused by deficient expression of monocarboxylic acid transporters (Additional file 2, Figure S2, [60]) or ketone body metabolizing enzymes (Figure 1B). Both neurons and astrocytes are in principle capable of using ketone bodies as metabolic fuels [61‐63]. In tumors, citric acid cycle metabolism is supposed to be intact and important for their ability to synthesize substrates for membranes, nucleic acids and proteins [64, 65]. In our experiments, the accumulation of HIF-1α observed after treatment with 3NPA also indicates an intact citric acid cycle in the glioma cell lines (Figure 3D). Other studies on malignant gliomas demonstrated reductions in electron transport chain activities [11], structural defects [10, 66] and DNA mutations [67] in mitochondria. Nonetheless, glioma cells have been shown to exhibit [10] and be able to modulate respiratory activity [68]. Therefore the basic requirements for energy yield from ketone bodies (enzyme equipment, intact citric acid cycle and respiratory chain) should be met. Alternatively, the 3-hydroxybutyrate offered could have been used primarily for the synthesis of lipids, but in this case a decrease in supernatant concentrations and perhaps an effect on proliferation would have been expected. However, the pure presence of a protein does not imply its proper function and we didn't analyze activities of ketone body metabolizing enzymes in the present study. In an analysis of various tumors and tissues of the nervous system, OXCT1 enzyme activity was found to be lower in glial tumors compared to normal human brain [15]. So enzyme activity could be the factor limiting ketone body metabolism [6, 18]. Taken together, we observed a lack of capability to degrade ketone bodies in the glioma cell lines. The deficiency in metabolizing ketone bodies might indeed represent a characteristic of malignant transformation, but the underlying defect remains unclear.

A reduction of tumor growth under conditions of caloric restriction and/or weight loss has repeatedly been shown in glioma models [17, 18]. However, we rarely observe chemotherapy-associated (unintended) weight loss in brain tumor patients and a non-calorie-restricted ketogenic diet might be more easily realized than calorie restriction. We therefore performed an in vivo experiment using an unrestricted ketogenic diet. The diet was well accepted, and no significant differences developed in body weight between the two diet groups (Figure 5A). Further, glucose concentrations and IGF-1 levels did not differ substantially between the groups (Figure 5B,Table 4). These results are consistent with prior analyses showing no decline in blood glucose concentrations when a ketogenic diet was administered in unrestricted amounts [18, 31, 69]. According to other studies [17, 69, 70], a reduction in circulating IGF-1 levels would have been expected only under conditions of caloric restriction. Thus, the stable values of body mass, glucose and IGF-1 help to distinguish between possible starvation-associated effects and other metabolism-specific effects of the ketogenic diet. Finally, the ketogenic diet alone did not influence tumor growth in the glioma model used (Figure 6A); tumor histopathology and metabolic mapping revealed no differences between mice fed the ketogenic diet and control animals (Figure 7). Although these results are consistent with those obtained in other syngeneic (CT-2A) and xenogeneic (U87MG) glioma mouse models [18], they contrast with the findings of Otto et al. where a diet identical to the one used in our in vivo experiment significantly decreased the growth of subcutaneously implanted tumors of the gastric adenocarcinoma cell line 23132/87 [31]. The different results of the latter and our study most likely reflect intrinsic properties of the cell lines used. In contrast to the gastric adenocarcinoma cell line, LNT-229 human glioma cells usually form quite homogeneous non-necrotic tumors. Necrotic tumors might be more susceptible to dietary restriction, as the necrotic areas in those tumors reflect the already limited supply of nutrients and oxygen. Since hypoxic tumor cells particularly depend on glucose availability [37], limiting carbohydrates might be effective in these tumors. The downregulation of ketone body metabolizing enzymes under hypoxic conditions observed in our cell lines (Figure 3A) suggests an additional disadvantage of this hypoxic tumor fraction. Another possible explanation may be a differential energy supply of subcutaneous and intracranial tumors. The mouse brain cells could have adapted to metabolize ketone bodies, leaving enough glucose to meet the xenografts' energy requirements. Using a different unrestricted ketogenic diet and the GL261 syngeneic intracranial glioma mouse model, Stafford et al. found a reduced rate of tumor growth and prolonged survival [71]. Likewise, GL261 tumors display necrotic zones [72]. Obviously, the impact of ketosis and hence the results of such studies depend on the model applied.

In summary, our results suggest a deficiency of glioma cell lines to metabolize ketone bodies in vitro, supporting the possibility of targeting tumor energy metabolism by a ketogenic diet. However, an unrestricted ketogenic diet was not effective as a monotherapy in the xenograft model applied. A combination of a ketogenic diet with strategies inhibiting glycolysis or interfering with the tumor's energy supply, such as vascular disrupting or antiangiogenic agents [73], might result in synergistic antitumor effects and is worth further investigation.