Urinary volatile metabolites of amygdala-kindled mice reveal novel biomarkers associated with temporal lobe epilepsy

Epilepsy is a chronic neurological disorder affecting mammals, including humans. Uncontrolled epilepsy is associated with poor quality of life, accidents, and sudden death. In particular, temporal lobe epilepsy (TLE) is the most common type of pharmacoresistant epilepsy, which easily gets out of control in human adults. The aim of this study was to profile urinary volatile organic compounds (VOCs) in a mouse model of TLE using solid-phase microextraction (SPME) gas chromatography mass spectrometry (GC-MS). Thirteen urinary VOCs exhibited differential abundance between epileptic and control mice, and the corresponding areas under the receiver operating characteristic (ROC) curve were greater than 0.8. Principal component analysis (PCA) based on these 13 VOCs separated epileptic from sham operated-mice, suggesting that all these 13 VOCs are epilepsy biomarkers. Promax rotation and dendrogram analysis concordantly separated the 13 VOCs into three groups. Stepwise linear discriminant analysis extracted methanethiol; disulfide, dimethyl; and 2-butanone as predictors. Based on known metabolic systems, the results suggest that TLE induced by amygdala stimulation could affect both endogenous metabolites and the gut flora. Future work will elucidate the physiological meaning of the VOCs as end-products of metabolic networks and assess the impact of the metabolic background involved in development of TLE.


Results
Thirty-one mice including 16 epileptic and 15 sham-operated control mice were included in the study. Stimulation of the amygdala once a day increased the number of spikes and the duration of the afterdischage, and finally induced tonic-clonic seizures (Fig. 1). Urine was collected at age 13.5 to 18.5 weeks in amygdala kindled mice.
Determination of the urinary volatile profiles of epileptic mice by SPME GC-MS. Typical SPME GC-MS TIC chromatograms of urine samples from an epileptic and a sham-operated mice are shown in Fig. 2, and indicate that very similar VOC profiles were obtained from the two groups of samples.
One hundred thirty-five metabolites were detected in both epileptic and sham-operated mice using GC-MS (Shimadzu QP-2010 Ultra and TQ-8040) (Table S1), including a variety of chemical structures, such as aldehyde, ketones, nitrogen compounds, terpenes, acids, alcohols, benzene derivatives, furan, sulphur, etc. XCMS extracted 24 VOCs showing differential fragment ion m/z values between the two groups of samples (Table 1). Next, the fragment ion m/z values of the VOCs with the largest area within each fragmentation pattern in the 24 VOCs identify the optimal cut-off value using the absolute area of each ion peak shown in Table 1 (data not shown). The area under the ROC curve (AUC), and sensitivity, specificity, and accuracy at the cut-off point were calculated to evaluate the discriminatory power of the potential biomarkers in tonic-clonic seizures ( Table 2). The statistical analysis of each individual compound significantly different between the groups revealed excellent predictive power, as shown by the AUC values: the AUC of disulfide, dimethyl was 0.9091 (95% confidence interval (CI) 0.7858 to 1.032), with an accuracy of 0.8571 (sensitivity = 0.8182, specificity = 0.9000). The AUC of RI1227 was 0.9091 (95% CI 0.7532 to 1.065) with an accuracy of 0.9048 (sensitivity = 0.9091, specificity = 0.9000). The AUC for 2,3,5-trithiahexane (disulfide, methyl (methylthio) methyl) was 0.9182 (95% CI 0.7635 to 1.073) with an accuracy of 0.9524 (sensitivity = 0.9091, specificity = 1.0000). On the other hand, the sensitivity for 2-butanone was 1, showing no false negatives, while the specificity for 2,3,5-Trithiahexane was 1, showing no false positives. The AUCs for 14 VOCs (the exception being 3,4-dehydro-exo-brevicomin (7-Exo-ethyl-5-methyl-6,8-dioxabicyclo[3.2.1]oct-3-ene)) were greater than 0.8, indicating high potential as prospective epilepsy biomarkers. Additionally, 11 out of 15 species of VOCs have been detected in human specimens (indicated by a hash in Table 2, http://www.hmdb.ca), the exceptions being methane, nitro-, 3,4-dehydro-exo-brevicomin, RI1227, and RI1449.
Next, as 15 VOCs is a large numbers of biomarkers, we proceeded to narrow their number. Specifically, we tried to reduce the number of variables using principal component analysis (PCA), classify the VOCs by producing a dendrogram, and narrow their number using linear discriminant analysis.
Principal component analysis and dendrogram analysis. The 15 VOCs differentially expressed in the urine of kindled vs. sham-operated mice (Table 1) and the 13 VOCs obtained after removing RI 1227 and RI 1449 (unknown compounds not found in human specimens) were analyzed by PCA using the absolute area of each ion peak. A separation trend was revealed in the three-dimensional PCA score plot using all 15 potential biomarker Desorption was performed at 240 °C for 10 min. The injection was pulsed splitless (closed for 3 min). Temperature programming consisted of an initial temperature 40 °C for 10 min, followed by an increase of 5 °C/min to 240 °C, and a 10 min hold at the final temperature. Numbers indicate the following metabolites, which showed similarity indices above 85%: (1) carbon dioxide/carbamic acid, monoammonium salt/dlalanyl-l-alanine; (2)  www.nature.com/scientificreports www.nature.com/scientificreports/ VOCs. The percentage of variance explained by the first three principal components was: PC1, 58.85% and 55.35% using 15 and 13 VOCs, respectively; PC2: 16.03% and 17.20%; PC3: 10.99% and 12.51%. Cumulatively, the first three components explained 85.87% and 85.06% of the variance, respectively. The principal component scores of kindled mice (Fig. 3A,B, red circles) are clearly separated from those of sham-operated mice (blue circles). This suggests that the urinary VOCs we identified are indeed able to distinguish kindled from sham-operated mice. The loadings of the 15 or 13 VOCs on the first three PCs are also shown in Fig. 3A,B (transparent circles), and are concentrated near the origin of the axes.
As the VOCs loadings of the 13 known VOCs were concentrated near the origin in the PCA scores plot, we applied promax rotation (kappa = 4) with Kaiser normalization, which converged in 5 iterations. After   Table 1. Identification of urinary volatile organic compounds (VOCs) by XCMS, showing differential levels between full-kindled and sham-operated mice. The urine of male mice was collected daily since they acquired epileptic seizures. The NIST 14 standard reference database 60 tentatively identified VOCs from peaks of total ion current (TIC) chromatogram as described in Table S1. VOCs in amygdala-stimulated kindled and shamoperated mice were analyzed with XCMS, resulting in 24 VOCs. Quantified ions were used to calculate the peak areas of the VOCs. a The area of an ion peak was used for quantification of urinary VOCs. b Retention indices of VOCs by the InertCap PureWAX column. c Retention indices of VOCs by the DB-1 column (n.d., "not detected"). d The similarity indices (SI) show the similarity with the mass spectrum from the NIST 14 standard reference database. VOCs with SI greater than 85 were described. e Mann Whitney U-test (two-tailed).
f There was no identical pattern of m/z with the authentic compounds and the retention index of the literature data exhibited in the NIST library, while the maximum similarity index with the MS spectrum was above 85%. g Urinary creatinine concentrations were determined by LabAssay TM Creatinine colorimetry kit (Wako Pure Chemical industries, Ltd. Osaka), based on the Jaffe method 59 . † VOCs were confirmed by identification with commercial standard references. #, ## Hashes indicate that the retention index of the VOCs calculated with retention time in the polar (InertCap PureWAX) # and non-polar (DB-1) ## columns were identical with the literature data, respectively.
Linear discriminant analysis. Finally, linear discriminant analysis was performed, as the Box's M test suggested homogeneity of the covariance matrices (F (6, 2540) = 0.207, p = 0.975). The 13 known VOCs were used in the analysis and the stepwise method using Wilk's lambda was applied to automatically select the best variables, with maximum value of the F probability for retention set at 0.05 and minimum for deletion at 0.10, resulting in an eigenvalue of 3.650 and canonical correlation of 0.886 (chi squared = 26.897, degrees of freedom (df) = 3, p = 6e-6). The standardized canonical discriminant function coefficients were 0.772 for methanethiol, −0.882 for 2-butanone, and 0.677 for disulfide, dimethyl. The discriminant function found is expressed by: where the square brackets represent the absolute area of ion peak m/z for each VOC. Using this function, the discriminant scores of kindled (black circles) and sham-operated (white circles) mice were calculated and are shown in Fig. 4. Finally, mice suffering from epilepsy were distinguished from controls by using the best biomarkers, i.e. methanethiol, disulfide, dimethyl, and 2-butanone, in which 100% of original grouped cases was correctly classified and 95.2% of cross-validated grouped cases was correctly classified (Fig. 4). Notably, each of these three VOCs belonged to a different group among those derived from the promax rotation (Fig. 3C) and the dendrogram analysis (Fig. 3D).

Discussion
It is known that the components of VOCs that are excreted from the human body reflect the metabolic condition of the individual. It has been suggested that VOCs could be useful in the olfactory diagnosis of several disorders 23 , including infectious diseases, inherited disorders of metabolism, and lung cancer 24 Table 2. ROC evaluations of urinary organic compounds (VOCs) showing differential levels between kindled (n = 11) and sham-operated (n = 10) mice. The absolute values of fifteen VOCs showing significant differences between kindled and sham-operated mice (p < 0.05) as shown in Table 1 were evaluated by ROC analysis. www.nature.com/scientificreports www.nature.com/scientificreports/ proposed the presence of seizure-alertness in dogs, who might recognize pre-ictal human behavioural changes, changes in heart rate, and olfactory cues, while Corne et al. 26 suggested that human prostate cancer could be detected by dogs sniffing the patient's urine. Based on the suggestions of these reports, we hypothesized the possibility that urinary VOCs in human epilepsy might be detectable by olfactory cues. On the other hand, in previous our study with a TLE model using amygdala-kindled mice, the expression of growth hormone, which is, main hormone involved in lipid metabolism, was up-regulated along the neural circuits 27 . A ketogenic diet, high in fat, improves seizures 4,5 in humans, suggesting that extracranial lipid metabolism is associated with convulsions seen in epileptic seizures. Hence, we formulated a hypothesis that the development of epilepsy is correlated with differences in organic compounds that originate from lipid metabolism, following development of epilepsy. As organic compounds that originate from lipid metabolism are highly volatile, we screened urinary VOCs in amygdala-kindled mice. We identified 15 types of VOCs, including two unknown-VOCs showing differential levels in the urine of amygdala-kindled mice, with AUCs above 0.8 (Table 2). Four VOCs, i.e., 2-buanone, 2-pentanone, and 2-heptanone of methyl ketones and 3,4-dehydro-exo-brevicomin are formed VOCs (B) in the urine of amygdala-kindled and sham-operated mice. PC1 to PC3 scores of kindled mice (red circles), sham operated mice (blue circles), and PC1 to PC3 standardized scoring coefficients (black circles) in the principal component score coefficient matrix of each VOC were represented as the X-, Y-, and Z-axis, respectively. Promax rotation with Kaiser normalization was applied to the PCA with 13 VOCs, and components 1 to 3 of the pattern matrix of each VOC were represented as the X-, Y-, and Z-axis, respectively (C). A dendrogram of 13 VOCs was derived with the Ward method from the component scores 1 to 6 of each VOC (D). The vertical axis shows the number described in Table 1 www.nature.com/scientificreports www.nature.com/scientificreports/ from fatty acids 28,29 (Fig. 5B), showing differences following kindled-seizures. 2-butanone, 2-pentanone, and 3,4-dehydro-exo-brevicomin increased 1.31, 1.99, and 1.39 times, respectively, and 2-heptanone decreased 0.55 times. This suggests that TLE induced changes in lipid metabolism, resulting in the differential urinary VOCs. On the other hand, 4 VOCs including sulphur, 2 VOCs of nitrogen compounds, and aromatic VOCs including 2-acetyl-1-pyrroline, 2-acetylpyrrole, and acetophenone were also detected as biomarkers of TLE in mice (Fig. 3). This suggests that epileptic seizures also induced other metabolic changes.
Promax rotation and dendrogram analysis separated the 13 VOCs into 3 groups (Fig. 3C,D). VOCs were classified based on differential abundance between epileptic and sham-operated mice. VOCs with increased levels, such as 2-butanone, 2-pentanone, and 3,4-dehydro-exo-brevicomin, in kindled-mice were categorized to one group. Among the decreased VOCs, 2 pyrrole and 2 nitrogen compounds were categorized as the second and third groups, respectively. On the other hand, 4 VOCs including sulphur were divided to two groups, proposing that differential metabolic system works between methanthiol/2,3,5-trithiahexane and disulfide, dimethyl/dimethyl trisulfide in mice.
The findings of TLE-responsive urinary VOCs were based on SPME collection and GC-MS analyses. As we focused on urinary VOCs that originated from lipid metabolism, we first selected a divinylbenzene/carboxen/ polydimethylsiloxane (DVB/CAR/PDMS) fibre, which is highly efficient for the collection of C3-C20 volatile and semi-volatile compounds. The extraction time was 60 min as described by Hanai et al. 24 . On the other hand, we used different temperature conditions (37 °C, 45 °C and 60 °C) below 60 °C, because of the possibility of degradation of the components at higher temperatures could inhibit the efficiency of the volatile VOCs. As a result, each TIC peak intensity increased in proportion to an increase in temperature. The number of TIC peaks were 109 at 37 °C, 143 at 45 °C, and 144 at 60 °C (Fig. S1). As there was little difference in the data obtained at 45 °C and 60 °C, we chose the data obtained at 45 °C.
Since we screened urinary VOCs in both amygdala kindling and sham operated-mice, in this study, we were unable to select an internal standard for each VOC for the SPME collection and GC-MS analysis. Hence, we determined the recovery rate of the SPME collection in C57Bl/6j urine (200 μL) including 100 ng of p-bromofluorobenzene (standard solution 021-12041, Wako Pure Chemical industries, Ltd.) against a liquid injection (1 μL) of 100 ng of p-bromofluorobenzene in GC-MS (Fig. S2). There was little difference of absolute area (m/z 174) between the liquid injection and SPME collection, as shown by the complete recovery by SPME collection (104%). We injected p-bromofluorobenzene collected by SPME at the start and the end of the multiple sample inspection using the multifunctional autosampler system AOC-6000 (162 wells for samples) in every GC-MS analysis and obtained the same absolute areas (m/z 174) between the start and end. Hence, the GC-MS was run once using the multifunctional autosampler system, the absolute area of each m/z in individual urine samples were analysed, and VOCs were compared semi-quantitatively, as shown in Table 1. Quantitative methods for urinary VOCs, which were determined as biomarkers for mice TLE, should be developed using internal standards, in the near future.
Next, we investigated the metabolic pathways to which the 13 VOCs are associated and how they might be associated to epilepsy and other phenotypes. First, the volatile sulfur compounds (VSCs) methanethiol; disulfide, dimethyl; dimethyl trisulfide; and 2,3,5-trithiahexane decreased in the urine of mice following kindled seizures. Hydrogen sulfide (H 2 S) and methanethiol are naturally formed in mammalian tissues 30,31 and by microbiota in the intestinal tract 32 (Fig. 5A). In mammalian tissues, Kolluru et al. 31 summarized in a 2013 review that H 2 S can be produced by enzymatic and non-enzymatic pathways: Cystathionine gamma-lyase (CSE) in the vasculature and liver, cystathionine beta-synthase (CBS) in the brain, nervous system, and liver, and 3-mercaptopyruvate sulfurtransferase (3-MST) in the brain and vasculature function as enzymes; non-enzymatically, H 2 S is generated www.nature.com/scientificreports www.nature.com/scientificreports/ from glucose via glycolysis or from phosphogluconate via NADPH oxidase. In fact, glucose reacts with methionine, homocysteine and cysteine, leading to methanethiol and H 2 S. Additionally, methanethiol is also produced by thiol S-methyltransferase with H 2 S 30 and oxidizes spontaneously to disulfide, dimethyl 33 . In the gut flora, the review by Martínez-Cuesta et al. in 2013 32 , focusing on lactic acid bacteria in cheese microbiota, states that methanethiol is normally derived from methionine in the presence of pyridoxal phosphate and oxidizes to dimethyl disulfide and dimethyl trisulfide. On the other hand, the metabolism of 2,3,5-tritiahexane in either mammalian tissues or gut flora remains unknown. However, a study showed that male mice release urinary 2,3,5-tritiahexane as a pheromone 34 (Fig. 5C) and Spadone et al. 35 demonstrated in 2006 that the thermal degradation products of methionine and photolysis of dimethyl trisulfide lead to the exogenous production of 2,3,5-tritiahexane. Hence, the endogenous metabolic systems related with the methionine-homocysteine cycle might affect polysulfuration. If so, decreased level of 2,3,5-tritiahexane might be observed following epileptic seizures as the decreased polysulfuration (Fig. 5A).
Second, 2-acetyl-1-pyrroline, 2-acetylpyrrole, and trimethylamine also decreased in mice urine following kindled seizures (Fig. 5A). The first two normally originate from food 36 , while trimethylamine comes from the gut flora 37 . 2-acetyl-1-pyrroline is the most important aroma compound in rice and can easily oxidize to 2-acetylpyrrole at room temperature, so that strong correlations between 2-acetyl-1-pyrroline and 2-acetylpyrrole determine the aromatic varieties of rice 36 . The phospholipid PC is the most significant dietary source of choline and the enzyme betaine aldehyde dehydrogenase (Badh2) metabolizes betaine-aldehyde to forms betaine via choline, and inhibits the biosynthesis of 2-acetyl-1-pyrroline 38 . On the other hand, mutations of the human gene ALDH7A1 39 , the homolog of Badh2, cause pyridoxine-dependent epilepsy 40 . In amygdala-kindled epileptic mice, activation of ALDH7A1 might lower the stability of 2-acetyl-1-pyrroline and 2-acetylpyrrole. It has been reported that microbiota metabolizes PC and choline to form trimethylamine 37 , which functions as a pheromone in both females and males 33,41 (Fig. 5C).
Third, methane, nitro-also decreased in mice urine following kindled seizures (Fig. 5A). While methane, nitro-has not been detected in human blood, excretions such as urine and feces, or saliva, rats subcutaneously Enzymes and coenzymes are enclosed in squares. Blue and red arrows show the differential VOCs (in italics) after kindled seizures; pathways described by green lines, arrows, and squares are derived from non-mammalian organisms.
www.nature.com/scientificreports www.nature.com/scientificreports/ injected with methane, nitro-were used as a model of human histidinemia, showing increased levels of histidine in blood, urine, and cerebrospinal fluid, in which methane, nitro-is a histidine ammonia-lyase inhibitor (HAL) 42 . Furthermore, it is known that L-cysteine also inhibits HAL activity 43 . As the histidine degradation system includes the production of glutamate and GABA (Fig. 5A), modulation of HAL activity might affect the development and/or severity of seizures.
Fourth, 2-butanone and 2-pentanone increased and acetophenone and 2-heptanone decreased in the urine of mice following kindled seizures (Fig. 5B,C). These four VOCs have generally been described as flavor ingredients included in various foods and detected also in human excretory biospecimens, such as feces, saliva, and/or urine (Human Metabolome Database). Metabolically, while acetophenone is derived from L-phenylalanine in plants, such as Camellia sinensis (tea) 44 , 2-butanone, 2-pentanone, and 2-heptanone are methyl ketones derived by the fatty acid beta-oxidation network 45 and glycolysis reactions in bacteria 28,46 (Fig. 5B). Specifically, it was demonstrated that 2-butanone and 2-pentanone are produced from glucose by Klebsiella pneumoniae in gut flora 47 and from hexanoic acids by Penicillium roqueforti in blue cheese 48 , respectively.
Associations of these VOCs with epilepsy and anxiety have been previously reported 49,50 (Fig. 5C). Pettersson et al. 51 showed that 2-pentanone reduced COX-2 protein levels in cultured cells originating from cancer, and Silva et al. 52 showed that COX-2 mRNA and acetophenone decrease synchronously with β-glucan in vitro. COX-2 is an inducible enzyme expressed as an immediate early response gene, and is involved in the derivation of prostaglandins from arachidonic acid, and associated with inflammation and cancer. We have demonstrated that the kindling mice upregulated COX-2 expression in the brain immediately after 3 seizures as the onset of epilepsy 9 . On the other hand, the present amygdala-kindled mouse suffers from chronic temporal lobe epilepsy that causes daily seizures for about 40 days after the onset. Cavazos et al. 53 observed neuronal loss in the CA3 field of hippocampal, entorhinal cortex, and the rostral endopyriform nucleus after 30 seizures. Currently, urine carrying the four VOCs was collected from chronic TLE mice suffering from 4 to 40 seizures, and not in the period immediately following onset. Therefore, it is suggested that urine during development of chronic TLE, accompanied with structural changes in the mesial temporal lobe, was collected. Hence, increase of 2-pentanone and decrease of acetophenone might occur as a negative feedback of epileptic seizures. Furthermore, reports have shown that 2-butanone blocked status epilepticus induced by lithium-pilocarpine in rats 49 , and 2-heptanone enhanced the neural activity of the basal amygdala accompanied by the removal of the anterior olfactory epithelial organs 54 . The urine of mice suffering from chronic TLE exhibited increased 2-butanone and decreases 2-heptanone, which might suggest a negative feedback of chronic seizures. Hence, when epilepsy progresses chronically, 2-butanone and 2-pentanone might be produced as seizure-suppressants, possibly accompanied by a decrease of acetophenone and 2-heptanone that are associated with neuronal activity.
Fifth, four of the VOCs identified as urinary biomarkers of mouse epilepsy, namely trimethylamine (TMA), 2,3,5-trithiahexane (TTH), 2-heptanone, and 3,4-dehydro-exo-brevicomin (DEB) are pheromones (Fig. 5D). Trimethylamine, 2,3,5-trithiahexane, and 2-heptanone, which have been detected in both mouse and human urine, decreased following mouse kindled-seizures, while DEB which has not been detected in human urine, increased. The excretion of trimethylamine in female mice and dimethylamine in women have been reported to be affected by diurnal rhythms 33 and trimethylamine also functions as an attractant pheromone for mice and an aversive one for humans via trace amine-associated receptor 5 (TAAR5) 41 . 2,3,5-trithiahexane is excreted by males 34 and a candidate human receptor was identified as OR2T11 55 , however its function is uncertain. The 2-heptanone excreted by males functions as an attractant pheromone to females and as an alarm one to males 50,54 . On the other hand, DEB is also a male pheromone attractant to females 56 and seems to share similar pheromone activity with trimethylamine and 2-heptanone. Currently, the opposite tendencies of these pheromones, namely the increase of DEB and the decrease of trimethylamine and 2-heptanone, remain to be explained. However, Song et al. 29 suggested that DEB is synthesized with omega-3-decenoyl-CoA, produced by a limited beta-oxidation with palmitoleic acid (16:1) and oleic acid (18:1), in the beetle body fat. On the other hand, we observed an increase of palmitoleic acid and oleic acid in the plasma of kindled-mice (manuscript in preparation), which might necessarily lead to a DEB increase.
Taken together, the urinary differential VOCs we identified suggest that TLE induced by amygdala stimulation could induce metabolic changes both endogenous and in the gut flora. Our preliminary observation was that there were few differences in the dietary content of the kindled and sham operated-mice, who were fed with the help of a metabolic cage (KN-645, Natsume Seisakusho, Tokyo, Japan), which was 3.4314 ± 0.2351 g in kindled-mice (average ± S.E.M., n = 7) and 3.5886 + 0.1795 in sham operated-mice (n = 7) (p = 0.8747 in Mann Whitney U-test, two-way). It is thought that repetitive seizures did not induce malnutrition, which might be associated with the changing gut flora. Thus, further investigations are needed into the TLE-induced changes in the gut flora. Moreover, we determined m/z areas of 15 VOCs in the diet (MF 12 mm ϕ pellet, Oriental Yeast Co. Ltd., Tokyo, Japan) (Table S2). Table S2B shows that 7 VOCs were detected in the diet from 15 VOCs, in which trimethylamine; methanthiol; 2-butanone; disulfide, dimethyl; 2-heptanone; dimethyl trisulfide; and 2-acetylpyrrole were found abundantly in the diet of the mice, suggesting both possibilities that the VOCs might be metabolized from those adsorbed from the diet and produced by the metabolic system. The other VOCs that are not found in the diet, might be metabolized endogenously and/or by the gut flora, such as 2-pentanone; methane, nitro-; RI1227; 2-acetyl-1-pyrroline; 3,4-dehydro-exo-brevicomin; RI1449; acetophenone; and 2,3,5-trithiahexane. In future research, we will take into account our VOC results in investigating possible correlations induced by TLE between peripheral metabolic system and brain function.
Epileptic seizures lead to alteration in the blood 57 . For example, the blood ammonia level increases in convulsions, inducing acidosis, leading to a medical emergency 58 . These are considered as epilepsy-responsive symptoms for two reasons: (1) elevation of blood ammonia level occurs along with extensive muscle contractions, resulting in acidosis, (2) cardiopulmonary arrest or haemorrhagic shock causes acidosis, resulting in ammonia production by the red blood cells, leading to hyperammonemia in patients 58  www.nature.com/scientificreports www.nature.com/scientificreports/ creatine, lactate, hormones including prolactin, and creatinine kinase, etc, are also detected postictally. However, these products are detected in both non-epileptic and epileptic seizures 57 . The metabolic changes induced by TLE still remain unexplained. We were unable to isolate VOCs related to hyperammonemia in the present study. We should investigate endogenous metabolism in the blood of mice with TLE induced by amygdala stimulation and the linkage of excreted urinary VOCs with endogenous blood metabolites.
In conclusion, mesial TLE includes foci in the amygdala, hippocampus and surrounding cortex and exhibits common symptoms in many mammals including humans. Many species of mammals have been used as experimental models of TLE, which can be induced by amygdala and hippocampus kindling stimulations to clarify the mechanisms of developmental TLE. The present results suggest that urinary VOCs, detected by SPME GC-MS, can potentially be metabolic biomarkers of TLE in mice. The hypothesis that altered urinary VOCs profiles may be derived from specific metabolic cascades could lead to the identification of common biomarkers for human and animals, and thus deserves further investigation. In particular, urine sampling could represent a simple and safe alternative to more invasive procedures in children and domestic animals.
Methods www.nature.com/scientificreports www.nature.com/scientificreports/ (RI), calculated in relation to the retention time of n-alkanes series (Kovats indices, 1958), with the data in NIST Chemistry WebBook 60 . We used the multifunctional autosampler system AOC-6000 (Shimadzu Co.) for stable measurement of all samples. Metabolite concentration was determined by calculating the ratio of the ion peak area of volatiles and the peak area of the limiting diluted external standard.
Statistical analyses. The absolute area of each ion peak was indicated as mean ± standard error of the mean (S.E.M.), and compared between the two groups using the Mann-Whitney U-test (Table 1), with p-values ≤ 0.05 considered as statistically significant. To determine the ability of each VOC to separate epileptic mice from sham-operated mice, receiver operating characteristic (ROC) curves were constructed, which plot "sensitivity" against "1 -specificity" (GraphPad Prism 6.07 for Windows, GraphPad Software, La Jolla, CA, USA). The area under the ROC curve (AUC) was calculated to estimate the predictive power of these potential biomarkers in distinguishing epilepsy from sham-operated controls.
In parallel, exploratory data analysis using the volatile compounds showing significant differences between the groups was performed by principal component analysis (PCA) (IBM SPSS Statistics 25), to validate the predicted probabilities of belonging to each group. During autoscaling, principal components were extracted with absolute values of VOCs in the "Factor Analysis" menu of the SPSS, in which "Correlation matrix" in analysis, "Unrotated factor solution" in display and six factors in extraction were selected as the condition of PCA. The principal factor scores for individuals (variables) and the principal factor score coefficient matrix were saved using the regression method. Additionally, VOCs were grouped based on their factor loadings resulting from promax rotation with Kaiser normalization, and using a dendrogram built with the Ward method from the unrotated loadings of the first six components in the component matrix extracted using PCA without rotation. Finally, stepwise linear discriminant analysis was performed (IBM SPSS Statistics 25), as the Box's M test suggested homogeneity of covariance matrices (F (6, 2540) = 0.207, p = 0.975).