Breath profiles of children on ketogenic therapy

Ketogenic diets (KDs) were initially introduced to clinical practices as alimentary approaches with the aim to control drug-resistant epilepsies. Over the decades, a large and growing body of research has addressed the antiseizure effect of various KDs, and worked out KD-based dietary regimens, including their acting factors and modes of action. KDs have also appeared in weight loss therapies. Therapy control, particularly at initiation, happens through regular blood analysis and control of urine ketone levels. However, there is a lack of fast, reliable, and preferably non-invasive methods to accomplish this. The detection of exhaled breath constituents may offer a solution. The exhaled breath contains hundreds of volatile organic compounds (VOCs), which can be modified by diet. VOC detection technology has resulted in low-cost sensors that can facilitate the self-monitoring of patients in the future if reliable breath markers are available. Therefore, it is of interest to investigate the composition of exhaled breath in children on KDs. Twenty-two pediatric patients between 4 and 18 years of age were recruited in this study. Eleven of them received a KD and suffered from epilepsy, with the exception of one child, who was admitted to a weight-reduction therapy. The control group involved 11 patients with neurological disorders but not on KD. Breath volatiles were analyzed using gas chromatography mass spectrometry (GC-MS) after preconcentration of the analytes on needle traps (NTs). We found that the breath concentrations of a number of VOCs, namely acetaldehyde, acetone, 2-methylfuran, methyl-vinyl-ketone, and 2-pentanone were significantly elevated in the breath of children on a KD in comparison to their control counterparts. Interestingly, breath ethanol was lower in patients on a KD than in non-KD patients. Association studies revealed an interrelationship among (i) lipid parameters and ketone bodies, (ii) methacrolein, methyl-vinyl-ketone, and high-density lipoprotein, as well as (iii) methyl-vinyl-ketone, acetone, and 2-pentanone, thus raising the possibility of a common metabolic source. The duration of diet was positively and negatively associated with breath acetone and breath ethanol, respectively. Some of the changes were linked to β-oxidation, but there are uncertainties in regard to metabolic sources of other metabolites. Lipid peroxidation and alteration of intestinal microbial composition may also be involved in the changes of VOC profiles during KD. Since lipids used for metabolism during KD originate from external sources, the processes occurring cannot simply be compared to and deduced from changes appearing in starvation; however, lipid mobilization is also evident in starvation. To find reliable and sensitive VOC markers that are linked to the respective ketogenic regimen, further investigations are needed to reveal the metabolic background.


Introduction
Since the 1920s ketogenic diets (KDs) as alimentary supplements for controlling seizures have been a valuable approach in the treatment of epilepsies, particularly in children [1]. Over the intervening years a large body of data has been collected to explain the way(s) how a KD may exert its beneficial effect on seizure control. Hypotheses are divided into three classes by their modes of action. Class one suggests a direct anticonvulsant effect of ketone bodies, class two suggests a reduction of neuronal excitability by cerebral ketone body metabolism, and class three suggests a less direct effect of the KD upon seizure control [2,3]. However, despite scientific endeavors it is still unknown which factors this benefit of KD can be attributed to.
One candidate molecule responsible for the antiepileptic action of the KD is acetone [4]. Its possible role in seizure control was already raised in the 1930s [5]. At that time extensive research was made without leading to any appreciated mechanism of action [5]. Nonetheless, the possible anticonvulsant properties of acetone were further substantiated by the experiments undertaken with isopropanol in rats [5]. Since both isopropanol and acetone were effective anticonvulsants, the seizure-controlling effect of isopropanol was dedicated to acetone [5]. Although in experiments with seizure-susceptible mice acetone had the strongest anticonvulsant efficacy among ketone bodies, it remains unclear whether acetone itself or its metabolites possess antiseizure activity [6][7][8].
Support for the acetone hypothesis comes from a breath acetone investigation in relation to seizure control [9]. A linear relationship between plasma and breath acetone has been reported [10]. Since the classic KD (cKD) is a high fat, low carbohydrate, and low protein diet, the formation of other metabolites related to lipid metabolism can also be expected, and thus should be investigated.
Over the decades several types of ketogenic regimens, termed neuroketotherapeutics, have been determined to control refractory epilepsies [11]. The modified Atkins diet, often abbreviated as 'MAD', is one of these alternatives to the cKD, and is less restrictive in terms of calories [12]. MAD is effective and tolerable in refractory epilepsies [13].
The aim of the present study was to examine the effect of a KD upon the exhaled air profile of children. This is the first time that the exhaled breath of children on a KD was investigated for metabolites other than acetone (or isopropanol). In the study group, patients having epilepsy or being on weight reduction therapy were included.
Our results show that in addition to acetone other volatile organic compounds (VOCs) are also detectable, and concentrations also change during the course of the KD. Some of these changes are linked to β-oxidation, while the appearance of other metabolites is perhaps due to the alterations in intestinal microbial composition or lipid peroxidation (LPO) evolving as a consequence of ketogenic therapies. The appearance of such metabolites in breath raises the probability of finding indicators more sensitive than acetone for the effectiveness of the KD.

Materials
All compounds listed in table 1 were purchased as pure liquids from Sigma-Aldrich, Merck, and Chem-SampCo companies.

Patients
Twenty-two patients (13/9, m/w) between 4 and 18 years old were recruited in this study. From the ten patients with epilepsy, nine were treated with the MAD and one with the cKD. Additionally, one patient received the MAD in order to lose weight because of obesity. The control group involved 11 patients with neurological disorders but were not on a KD. The study was approved by the Ethical Committee of the Innsbruck Medical University (Study number: AN4315).

Blood samples
Blood samples were collected routinely by the attending physician for determination of blood glucose, liver parameters (GOT, GTP, γ-GT), kidney parameters (urea, creatinine), blood ketone concentrations (whole ketone bodies, acetoacetate, ß-hydroxy-butyrate), and drug levels. Blood samples were collected after overnight fasting in 9 of 11 patients in the KD group, and in seven of ten patients in the control group. Blood ketone concentrations were determined in the laboratory of the Pediatric Department of the Medical University of Innsbruck by a cyclic enzymatic test using photometric measuring of the rate of Thio-NADH production (Autokit Total Ketone Bodies; Wako Life Sciences, Mountain View, USA). All other parameters were measured in the central laboratory using standard procedures.
Breath sampling protocol Alveolar breath and indoor air samples were collected at the pediatric neurology clinic in the Pediatric Department of the Medical University of Innsbruck. As much as possible the same room was used for sampling to minimize the influence of indoor air contamination.
Nine patients in the KD group and seven in the control group gave samples after overnight fasting. The alveolar breath samples were collected manually into glass syringes (250 ml volume, Socorex Isba S.A, Ecublens, Switzerland). The glass syringe was capped with a three-way valve (Discofix, B. Braun Melsungen AG, Melsungen, Germany), and additionally connected via Teflon tube to a single-use mouthpiece (Intersurgical complete respiratory systems, Sankt Augustin, Deutschland) on which, with the help of a single-use Teflon-adapter, a CO 2 sensor (Phasein, Danderyd, Sweden) was mounted. During sampling the breath CO 2 content was monitored using the CO 2 sensor. During exhalation, syringes were filled with exhaled air if the absolute level of CO 2 in the breath exceeded 3%. Ambient air was collected in parallel (also in glass syringes). After sampling the syringes were locked with the three-way valve.
All samples were processed within 2 h. Before use, all syringes were cleaned, washed out with distilled water, and dried overnight at 100°C to remove any residual contaminants.
Sample preparation for gas chromatography mass spectrometry (GC-MS) measurements Extraction of VOCs was performed using stainless steel needles containing 2 cm Carbopack X and 1 cm Carboxen 1000 sorption materials, a so-called needle trap (NT, PAS Technology, Magdala, Germany). For VOC extraction, the syringe was pierced through a septum by the NT connected at the other end with an electronic mass flow controller (model F-201DV-RAD-11-V, Bronkhorst, Ruurlo, The Netherlands) via a Teflon tube. For the generation of sample flow, a pump (Vacuubrand, Wertheim, Germany) was placed at the end of the sampling system. The collected volume of breath sample was 200 ml, with a total flow of 8 ml min −1 through the NT, which was controlled by means of the mass flow controller. After extraction, NTs were cleaned by thermal heating at 300°C for 15 min.

GC-MS analysis
The sampled analytes were released from sorbents by direct thermal desorption of the NT at 290°C in splitless mode in the heated injector of the 7890A gas GC equipped with a 5975 C Inert XL mass selective detector (MSD) (both from Agilent Technologies, Waldbronn, Germany). The CombiPAL autosampler (CTC Analytics AG, Zwingen, Switzerland) was modified (by PAS Technology) for application of NTs allowing an automated sample processing.
MS analyses were performed in full scan mode, with a scan range of 20-200 amu. Ionization of the separated compounds was done by electron impact at 70 eV. Chromatographic data was acquired using the Agilent Chemstation Software (GC-MS Data Analysis from Agilent, Waldbronn, Germany), and the mass spectrum library NIST 2008 (Gatesburg, USA) was applied for identification. An RT-Q-Bond capillary column 30 m×0.32 mm×10 μm (Varian, Palo Alto, CA, USA) was used. The oven temperature program was as follows: initial 55°C for 0 min, then ramped 6°C min −1 up to 120°C and held for 1 min, then ramped 9°C min −1 up to 180°C and held for 1 min, then ramped 8°C min −1 up to 200°C and held for 1 min, and finally ramped 8°C min −1 up to 280°C and held for 16 min. This was conducted in a constant flow mode (helium at 2 ml min −1 ).

Calibrations
For determination of the retention time of compounds detected in room air and breath samples, gaseous standards were prepared by evaporation of liquid substances into glass bulbs. Each bulb (Supelco, Bellefonte, PA, USA) was cleaned with methanol (Sigma-Aldrich, Steinheim, Germany), dried at 85°C for at least 20 h, purged with clean nitrogen for at least 20 min, and subsequently evacuated using a vacuum pump (Vacuubrand, Wertheim, Germany) for 30 min. Liquid standards (0.5-1 μl according to desired concentration) were injected through a septum, using a GC syringe. After the evaporation of standards, the glass bulb was filled with nitrogen of purity 6.0 in order to equalize the pressure (to the ambient pressure). Then, the appropriate volume (μl) of vapor mixture was transferred using a gas-tight syringe (Hamilton, Bonaduz, Switzerland) into 250 ml glass syringes previously filled with 200 ml of nitrogen 6.0 (99.9999% purity). Detection limits were evaluated from the calibration curves using a t-distribution with 95% probability (table 1).

Data evaluation
Integration of chromatograms was done by means of Agilent MSD Productivity Chemstation Software (Agilent technologies, Santa Clara, California, 1999) and Breath View (Testversion 1.6 Oncotyrol 2014, Innsbruck, Austria). Mass spectra library NIST 2008 (Gatesburg, USA) was used for peak identification.

Statistical analysis
To compare the treated (MAD and cKD) versus control patient groups we used the Wilcoxon signed rank test as a non-parametric test to calculate the significance of the hypothesis because a normal distribution could not be determined owing to the small number of patients. In the case of a p value<0.05 mean values can be significantly differentiated, and the value is influenced by the KD.
Correlations between VOC concentrations in breath gas and other parameters were calculated using Spearman coefficients with a significance value of p<0.05.

Results
Basic characteristics of study and control groups Table 2 provides a statistical summary of background factors for subjects on and not on a KD. No differences between factors were found. The actual medication of the children is also shown.

Changes in ketone body levels
To control the efficacy of the regimen, the metabolic effects of a KD upon plasma ketone body levels were also determined, and are documented in table 3. As expected, the plasma concentrations of ketone bodies, either the separate or the total concentrations, significantly increased in the KD group in comparison to the non-KD group (table 3).
Ketogenic regimens have been reported to affect serum lipid profiles [14,15]. To ensure the revelation of correlations among variables, the lipid parameters (cholesterol, high-density lipoprotein (HDL), lowdensity lipoprotein (LDL), triglyceride) for patients enrolled in a KD were also determined. Their serum lipid profile was similar to that reported in the literature for ketogenic treatments (data not shown).

Measurement of VOCs in breath
The breath profile was examined in both groups. Altogether, 18 VOCs were selected and monitored (table 4). Neither 2-butanol nor hexanal were present at detectable amounts in the breath in either group. Furthermore, ethylacetate was only measurable in the exhaled air of patients on a KD, reaching approximately 0.1 ppb (n=11), which is very close to the detection limit (0.08 ppb) of the device. In table 4, the values for the other 15 VOCs are shown.
A significant rise in the amount of exhaled acetaldehyde, acetone, 2-methylfuran, methyl-vinylketone, and 2-pentanone was measured in the KD group. Interestingly, the breath ethanol level was lower in patients on KDs in comparison to controls. At the same time, no significantly different levels of isoprene, methacrolein, nonanal, octanal, pentane, 1-pentanol, or 1-propanol were detected (table 4).
As reported earlier, β-oxidation takes place in the breakdown of valproic acid, and 3-heptanone is an end-product of this process [16]. Since the KD also influences the rate of β-oxidation, the level of this metabolite in breath was determined (table 4). Valproic acid levels in plasma for KD and non-KD subjects were also measured, and we found a nonsignificant difference with values of 52.57±27.84 (mg l −1 ) (n=3) and 73.03±41.46 (mg l −1 ) (n=3), respectively. As expected, a strong correlation between the levels of 3-heptanone and valproic acid was seen (r=0.9293, p<0.001 ). In topiramate-treated cases, the values of topiramate were 3.7 (mg l −1 ) and 7.6 (mg l −1 ) in patients belonging to diet and non-diet groups, respectively.

Association among variables
To determine the interrelationships among variables, we assessed breath and serum variables pair-wise by linear regression analysis (table 5).
As expected, correlations were found among lipid parameters (cholesterol, HDL, LDL, ketone bodies) (table 5). Furthermore, the length of therapy and breath acetone showed a positive correlation, while the length of therapy and breath ethanol showed a negative correlation (table 5). Associations were revealed among methacrolein, methyl-vinyl-ketone, and HDL, as well as methyl-vinyl-ketone, acetone, and 2-pentanone, thus suggesting the possibility of a common metabolic source (table 5). However, this assumption has to be clarified in the future.

Discussion
The central issue in this research has been to identify VOCs that are intrinsic to ketogenic regimens, and can serve as biomarkers for treatment efficacy. A biomarker reflects biological processes and responses to therapeutic interventions in patients; therefore, the excretion of VOCs via breath is presented in this paper with regard to their appearance in the course of diet, along with a set of VOCs for which changes in their concentrations are linked to the respective ketogenic regimen.
The biochemical backgrounds of these changes are quite different. There are metabolites whose origins are not clearly identifiable at present. However, despite uncertainties the events can essentially be divided into two main groups of metabolic conversion: changes due to metabolic events inside the body that result from the introduction of a lipid-rich diet, and those seen as a consequence of the rebuilding of the structure of human gut microbiota.

Changes of metabolism during KD in humans
The KD involves a special composition of food. The cKD is a high-fat, low-carbohydrate, and low-protein diet that, in principle, mimics starvation, and is known for its beneficial effect on seizures since Biblical times (see Matthew 17:21). Over the decades, the cKD has been modified to avoid its side effects and starvation,  ). b Without any significant difference between the groups. c Six and four patients were free of any medication in the KD and non-KD groups, respectively. In addition, two patients received two or three kinds of medication in both groups. but has remained rich in lipids and restricted in carbohydrates [11,12,14]. The core biochemical change in starvation is an increase in β-oxidation accompanied with a decrease in glycolysis, leading to the elevation of plasma levels of all the three members of the ketone body family [17][18][19]. In general, plasma concentrations of acetoacetate, β-hydroxy-butyrate, and acetone increase, and exceed the mmolar range in starved humans [20]. In cKD-treated patients, the levels of these compounds reach a similar level [10,21]. Acetone concentration may exceed the level of 4 mM, and correlates with breath acetone [10]. Patients successfully treated with a cKD had elevated levels of acetone in their brains [22]. In our study, β-hydroxy-butyrate and acetoacetate concentrations were as high as 4.1 mM and 1.0 mM, respectively, with a concomitant increase in breath acetone and isopropanol levels (tables 3 and 4). In this sense, our findings are in accordance with the data reported in the literature [10,21]. The duration of the KD also correlated well with breath acetone level (table 5), which is in contrast to the finding of others [23]. However, this difference may be due to the fact that the cKD and MAD are not identical to each other, for example, in regards to the presence or absence of calorie restriction.
In the course of β-oxidation, every turn of the cycle shortens the carbon chain by two carbons, leading to the production of one molecule of acetyl-CoA, NADH+H + , and FADH 2 (figure 1). Nevertheless, there is another opportunity for the outflow from the cycle at the level of 3-ketoacyl-CoA (figure 1). The leakage of the cycle at this level may result in the formation of both 2-pentanone and 2-heptanone Table 4. Levels of VOCs in breath in ppbV. Data are presented in minimum (min), 25th percentile (Q1), median, 75th percentile (Q3), and maximum (max) values.

Diet group (ppbV)
Non-diet group (ppbV)   (figure 1). 2-Pentanone was well detectable in the breath of children of both groups, with a significant elevation in patients on a KD in comparison to controls (table 4); it also showed a positive correlation to breath acetoacetate, β-hydroxy-butyrate, and total ketone body levels, which were all related to β-oxidation (table 5). The duration of the KD was also correlated with 2-pentanone levels (table 5). Development of more sensitive detection methods may enable the monitoring of 2-heptanone, thus giving further evidence for the role of 3-ketoacyl-CoA. However, an alternative way to achieve 2-pentanone formation is through an incomplete β-oxidation of medium-chain fatty acids, probably in peroxisomes; this is a mechanism similar to that used in filamentous fungi [24].
In cholesterol synthesis, acetyl-CoA is a precursor molecule for mevalonate formation that is transformed into dimethylallyl pyrophosphate, a compound from which isoprene (2-methyl-1,3-butadiene) is produced [25]. In children, exhaled isoprene levels show an age-dependent character, whereas this correlation is not seen in adults [26][27][28]. In this study, an increase in exhaled isoprene levels was detected in patients on a KD compared to the those not on a KD, but it did not reach the level of significance (table 4). King and associates [29] assumed that isoprene is stored in muscles, and can be released fast during movement. Therefore, breath isoprene sampling analysis requires carefully rendered breath sampling, and patients should avoid movement, e.g. remain at rest for at least 5 min, and sitting before sampling. This criterion is difficult to achieve in the case of children. Despite the lack of significance, the trend in breath isoprene level might reflect a metabolic pressure built up by the rise of acetyl-CoA and NADH+H + (figure 2). The latter is known to slow down the TCA cycle, resulting in a reduced acetyl-CoA breakdown, thus creating a situation that may enhance the operation of the mevalonate pathway. This note is supported by the fact that isoprene may be the source of methyl-vinyl-ketone and methacrolein formation in the course of LPO by the hydroperoxyl pathway, and a specific type of oxygen free radicals is involved in the process [30,31]. Although the present study revealed a correlation between the amounts of methyl-vinyl-ketone and methacrolein (tables 4 and 5), any correlation to isoprene was not seen (data not shown).

LPO
LPO is the oxidative degradation of lipids. This process proceeds by a free radical chain reaction. A wide variety of compounds can be detected in the course of LPO, among others ketones, such as 2-pentanone and 3-pentanone, or alkanals, such as hexenal or octanal [30]. We detected compounds (e.g. pentane) that are related to LPO, while others also known as LPO products (e.g. hexanal) were undetectable.
There seems to be an agreement that LPO does not have a pathological role in the KD because LPO-related events are not mentioned among the late-onset complications [12,32]. Nevertheless, this optimistic opinion might have to be questioned. Although there are reports having found the KD protective against oxidative stress, only short-term investigations are available [33][34][35]. In clinical practice, KDs are longterm treatments, e.g. in our case the longest therapy had lasted 54 months. However, the studies from which the conclusion regarding safety has been drawn are comparatively short. Assuming that KDs are mainly used in the case of intractable seizures, one can suggest that a long-term treatment has to occur; therefore, from this point of view it would be relevant to see whether LPO contributes to side effects.
Although our present results do not fundamentally oppose the above view, they provide some reasonable doubts. An example is pentane. Breath pentane is used as an index to monitor LPO in animal models [36,37]. Its analysis in breath may provide an early, rapid, non-invasive, and real-time assessment of LPO [37]. In addition, its measurement is advised in clinical practice to assist in the treatment of LPO-associated disorders in preterm infants [38]. In our case, breath pentane levels increased in the KD group, almost reaching the level of significance (table 4). Perhaps a larger sample size would have provided a significant difference between groups.
Several arguments that may support the aforementioned skepticism, and direct attention to the possible vascular adverse effects of fat-rich regimens, should also be addressed. Firstly, note that the NADH+H + burden may also enhance LPO, in which Fe 2+ ions play a role [39]. Secondly, note that acetoacetate, but not β-hydroxy-butyrate, has been reported to cause an elevation in LPO in cultured human venous endothelial cells through an Fe 2+ iondependent oxygen free radical generation [40]. Thirdly, note that acetoacetate leads to methylglyoxal production via reactive oxygen species generation in the presence of molecular oxygen; in the reaction either myoglobin, or, less effectively, hemoglobin takes part [41,42]. Fourthly, note that the KD leads to a higher arterial stiffness (an early marker of vascular damage), and elevated cholesterol and triglyceride levels in treated young subjects in comparison to controls [43]. Fifthly, note that seizures themselves rise LPO [44]. Finally, note that an impairment of the glutathione system has been reported in patients with epilepsy independent from seizure frequency [45]. Therefore, a modest speculation is appropriate here. A KD for patients, seizure-free or not, may open the way for LPO with a proposed mechanism as depicted in figure 2.
Metabolites in which their production can be related to changes connected to rebuilding gut flora in the course of diets Since ethanol is oxidized to acetaldehyde, a reaction catalyzed by alcohol dehydrogenase (ALD), one would expect the changes in the levels of these compounds in breath to run parallel. However, this is obviously not  (table 4).
One possible explanation of these findings is based on two arguments. First is the auto-brewery syndrome, which is a well-known phenomenon leading to ethanol production by gut flora [46]. Ethanol generated in the gut is mainly converted to acetaldehyde to produce NADH+H + by bacterial ALD b (figure 3). Both ethanol and acetaldehyde can be secreted into the gut lumen and taken up by enterocytes. In human tissues, a large amount of NADH+H + is generated as a result of ketogenic therapies and inhibits ethanolacetaldehyde and acetaldehyde-acetate conversion by human ALD h and aldehyde dehydrogenase, respectively; however, this makes the reverse reaction thermodynamically favorable [46]. At the same time, acetyl-CoA levels are also elevated, which is an obstacle to the further oxidation of acetaldehyde. This is the case for enterocytes as well ( figure 3). In the case of non-KD subjects, ethanol is fairly oxidized to aldehyde. This argument explains why changes moved in different directions.
The second argument is the rebuilding of gut flora. The relationship between humans and their intestinal microbial community is complex and mutually favorable under physiological conditions. The gut ecosystem provides beneficial products (e.g. vitamins) for the host, while the host grants a suitable environment for microbiota [47,48]. The structure of the gut microbiome varies as a function of adaptation to the intestinal environment depending on diet, diseases, or antibiotics taken [47][48][49][50]. Ethanol production of gut flora depends on microorganisms and nutritional conditions [51]. The cKD reduces total intestinal flora, which results in changes in the metabolites produced [52]. Nevertheless, this is a field that warrants further investigation.

Effect of food
There are compounds, e.g. furan and its metabolite 2-methylfuran, that are present in food [53]. It is, however, a question to what extent food can be responsible for elevated levels of 2-methylfuran in breath. Despite this uncertainty, 2-methylfuran was detected at a significantly elevated level in the KD group in comparison to the non-KD group (table 4). The reason could be its higher intake from food, but this suggestion needs further verification.

Limitations of the study
In this study the number of patients enrolled was relatively low. This may explain why the p value did not reach the level of significance in some cases.
A further limitation of this study is to what degree group unification was successfully undertaken. Two points need to be mentioned here. Firstly, the age range of children was wide (4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15)(16)(17)(18), and this made the interpretation difficult because of the broader concentration range of the exhaled volatiles. Secondly, exploring the family environment did not extend to factors that might have potential importance for the study, for example, the smoking of family members. Methyl-vinyl-ketone in the breath of KD group patients was significantly elevated in comparison to controls (table 3). However, methyl-vinyl-ketone is not only a major oxidation product of isoprene [31], but is also among the hundreds of metabolites found in cigarette smoke [54]. Although isoprene, methacrolein, and methyl-vinyl-ketone levels changed in similar ways, a child living with smoking parents may present higher levels than one living with non-smokers. This rule also applies to other compounds abundant in urban environments, e.g. furan [54].
It is also rational to suggest that different neurological conditions in the control group (non-diet group) might have an effect on the metabolic network. This is part of the problem of group unification, which is a function of patient recruitment and sample size. Nevertheless, we are short of literature data that either supports or opposes this suggestion.

Conclusions
In conclusion, several volatile metabolites were detected in the breath of children. Changes in the levels of metabolites could be explained in the majority of cases, and were related to biochemical changes caused by ketogenic therapy. Figure 4 depicts the basic biochemical mechanisms and changes that were observed. It is, however, a question of future research which mechanism explains the clinical findings and which metabolite can be used as a biomarker in ketogenic therapies As it now looks, besides the acetone/isopropanol pair, the detection of 2-pentanone and the monitoring of acetaldehyde and ethanol together may be more promising as a marker due to the fact that their concentrations change in opposite directions, thus making it possible to distinguish the ketotic state occurring in a KD from other states, such as alcoholism and probably diabetes mellitus.
Finally, one obvious and crucial question has to be addressed in future studies. Although both starvation and ketogenic regimens are characterized by high levels of lipids and ketone bodies, the fundamental difference between them is that in the former state the body mobilizes its own metabolic sources (switching from mainly glucose to lipid-based energy provision [55]), while in the latter case lipids arise from an external source. If this difference is significant then the metabolic and cellular events cannot be one and the same.