{"paper_id":"0622b967-4907-41d3-80a2-bd8dbee16ea6","body_text":"Alterations in serum metabolomics predict drug-resistant epilepsy | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Alterations in serum metabolomics predict drug-resistant epilepsy Wenzhong Kang, Qingrong Han, Min Chen, Longxing Xue, Zhihua Zhao, and 5 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7173411/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 03 Dec, 2025 Read the published version in Scientific Reports → Version 1 posted 10 You are reading this latest preprint version Abstract Drug-resistant epilepsy (DRE) lacks a reliable early warning system. This prospective study developed an early predictive model for DRE based on baseline serum metabolite levels. Serum samples were collected from 151 prospectively recruited patients with epilepsy. Patients were categorized into the DRE and non-drug-refractory epilepsy (NDRE) groups after a four-year follow up. After propensity score matching of baseline data, including age and sex, 33 patients with DRE and 89 with NDRE were recruited and split into training and test sets at random in a 2:1 ratio. Nontargeted metabolomics of the training set identified 109 significantly altered metabolites, primarily those involved in dysregulated lipid metabolism. Compared with the NDRE group, 22 metabolites were upregulated and 87 were downregulated in the DRE group. Pathway enrichment analysis highlighted perturbations in choline metabolism in cancer and sphingolipid, glycerophospholipid, one-carbon pool by folate, and cholesterol metabolism. A support vector machine model incorporating 11 metabolites and one clinical characteristic achieved a cross-validated area under the curve (AUC) of 0.810 and an independent test set AUC of 0.794. This study provides a non-invasive, serum-based objective tool to identify the potential population of patients with DRE with good sensitivity and specificity and guide targeted metabolic therapies. Health sciences/Biomarkers Biological sciences/Cancer Health sciences/Diseases Health sciences/Medical research epilepsy drug-refractory epilepsy serum metabolomics predictive model Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Introduction Epilepsy is a prevalent chronic neurological disorder distinguished by the occurrence of unprovoked seizures and has a global prevalence of 0.5–1% in developed countries and significantly higher rates in developing countries [ 1 ]. Although medication can control seizures in most patients, some patients still experience uncontrollable seizures due to drug resistance. In 2010, the International League Against Epilepsy (ILAE) defined drug-resistant epilepsy (DRE) as the failure of two appropriately chosen and tolerated anti-seizure medication (ASM) regimens (monotherapy or polytherapy) to achieve freedom from seizure for more than one year [ 2 ]. Systematic reviews have estimated the prevalence at 30%, with an annual incidence of 15% [ 3 ]. Patients with DRE have a higher risk of premature mortality, neuropsychological impairment, and a significant reduction in quality of life [ 4 ]. Consequently, it is crucial to identify DRE in patients as soon as possible. The mechanisms behind drug resistance in epilepsy are still unknown. Current diagnostic paradigms remain reactive and rely on the outcome of medical treatment rather than predictive biomarkers, which hinders early individualization and optimization of treatment for patients diagnosed with DRE [ 5 , 6 ]. Several models and predictors have been proposed to predict DRE, including clinical and biochemical markers such as age at onset, abnormal electroencephalogram (EEG), seizure type, initial frequency of seizures, family history of epilepsy, aberrant neuroimaging, levels of high mobility group box 1, and SCN1A polymorphisms [ 7 ]. However, there are currently no molecular biomarkers able to identify possible DRE populations early on with good sensitivity and specificity. The measurement of small molecule concentrations in blood, urine, cerebrospinal fluid, and other biological matrices is known as metabolomics [ 8 , 9 ]. It offers data for ensuing \"omics\" technologies like proteomics and genomics. Metabolomics has become a significant method for the identification of disease biomarkers, characterized by its efficiency and high-throughput precision. It not only unravels intricate biological mechanisms but also serves as a reliable way to monitor therapeutic effects. Metabolomics is widely recognized as an effective approach to disease diagnosis, classification, staging, treatment planning, and prognostic evaluation [ 10 ]. In 2017, the ILAE classified “metabolic etiology” as one of six major etiological categories of epilepsy [ 11 ]. There are several subtypes of metabolic epilepsy, such as mitochondrial diseases, folinic acid-responsive seizures, and cerebral folate insufficiency [ 12 ]. Basic research and translational studies have shown that the relationship between cerebral metabolism and seizures is complicated and bi-directional and forms a harmful cycle that exacerbates the negative effects of seizures. Metabolic moleculars and enzymes have turned into appealing biological targets for preventing epileptic seizures and fostering recuperation, and metabolism-based therapy methods, such as high-fat antiepileptic ketogenic diets, have progressively gained traction [ 13 ]. Therefore, this study aimed to elucidate the differences in baseline serum metabolomics between DRE and non-drug-refractory epilepsy (NDRE) through a prospective cohort study. This study aimed to clarify the material basis for the occurrence and progression of DRE from a metabolic perspective, thereby providing a biological foundation and novel insights for further investigation of the pathophysiological mechanisms underlying DRE. Additionally, a predictive model was developed utilizing the clinical traits and potential metabolites identified in the training cohort and subsequently confirmed in the test cohort. Results Participants’ clinical features and basic information The study flow chart is displayed in Fig. 1 . The clinical characteristics and demographic information of the study population are presented in Table 1 . Other than the number of seizures prior to diagnosis, which also did not exhibit a statistically significant difference ( P > 0.05), there were no statistically significant differences between the DRE and NDRE groups ( P > 0.05). In the training cohort, the DRE group exhibited more frequent seizures before diagnosis than the NDRE group ( P = 0.004). Additionally, the NDRE group exhibited greater levels of total cholesterol than the DRE group within the test cohort (3.97 vs. 3.44 mmol/L, P = 0.011), but they were still within the normal physiological range. Table 1 The study cohort's demographic data and clinical features. Item Train cohort Test cohort DRE (n = 26) NDRE (n = 55) P - value DRE (n = 7) NDRE (n = 34) P - value Female sex, n (%) 11 (42.31) 21 (38.18) 0.723 4 (57.14) 17 (50.00) 1.000 Weight (kg), mean (SD) 58.2 (17.8) 57.7 (17.3) 0.905 59.2 (18.2) 59.2 (18.2) 0.077 Age (years), n (%) 0.570 0.910 Childhood (0–10) 4 (15.38) 7 (12.73) 1 (14.29) 7 (20.59) Adolescent (11–17) 6 (23.08) 19 (34.55) 2 (28.57) 8 (23.53) Adult age (≥ 18) 16 (61.54) 29 (52.73) 4 (57.14) 19 (55.88) Age at onset of epilepsy (years), n (%) 0.874 0.057 Childhood (0–10) 8(30.77) 14(25.45) 4(57.14) 10(29.41) Adolescent (11–17) 8(30.77) 19(34.55) 3(42.86) 12(35.29) Adult age (≥ 18) 10(38.46) 22(40.00) 0(0.00) 12(35.29) Number of seizures before diagnosis, n (%) 0.004 0.194 <4 10 (38.46) 39 (70.91) 4 (57.14) 21 (61.76) 4–10 2 (7.69) 8 (14.55) 0 (0.00) 6 (17.65) 10–100 12 (46.15) 7 (12.73) 3 (42.86) 7 (20.59) > 100 2 (7.69) 1 (1.82) 0 (0.00) 0 (0.00) MRI, n (%) 0.310 0.060 Abnormal 14 (53.85) 23 (41.82) 6 (57.14) 13 (38.24) Normal 12 (46.15) 32 (58.18) 1 (42.86) 21 (61.76) Electroencephalogram, n (%) 0.053 0.615 Abnormal 20 (76.92) 30 (54.55) 4 (57.14) 13 (38.24) Normal 6 (23.08) 25 (45.45) 3 (42.86) 21 (61.76) WBC (x10*9), mean (SD) 5.70 (1.38) 6.04 (1.67) 0.382 5.92 (1.14) 5.44 (1.50) 0.434 RBC (x10*9), mean (SD) 4.37 (0.46) 4.27(0.58) 0.448 4.04 (0.58) 4.28 (0.52) 0.287 Hb (g/L), mean (SD) 131 (14) 131 (14) 0.911 135 (14) 132 (14) 0.629 PLT (x10*6), mean (SD) 222 (82) 232 (75) 0.593 209 (42) 231 (71) 0.448 ALB (g/L), mean (SD) 44.3 (5.0) 43.2 (5.9) 0.424 42.0 (2.3) 43.3 (6.6) 0.615 TCHO (mmol/L), mean (SD) 3.82 (0.40) 3.79 (0.43) 0.720 3.44 (0.39) 3.97 (0.49) 0.011 TG (mmol/L), median (IQR) 0.99 (0.64,1.37) 0.72 (0.56,1.28) 0.230 1.15 (0.74,1.33) 0.74 (0.59,1.24) 0.100 TBIL (umol/L) median (IQR) 6.3 (4.4,8.5) 6.2 (4.7,8.3) 0.686 5.3 (5.1,6.3) 6.0 (4.5,7.4) 0.615 While continuous variables are displayed as mean (standard deviation, SD) or median (interquartile range, IQR), categorical variables are displayed as percentages [n, (%)]. MRI, magnetic resonance imaging; WBC, white blood cells; RBC, red blood cells; Hb, hemoglobin; PLT, platelet; ALB, albumin; TCHO, total cholesterol; TG, triglyceride; TBIL, total bilirubin; DRE, drug-resistant epilepsy; NDRE, non-drug-refractory epilepsy. Metabolomic profiling and data quality assessment Non-targeted metabolomic analysis was performed on the serum samples of 81 patients with epilepsy in the training set (55 patients with NDRE and 26 with DRE). Data pre-processing, including peak alignment and filtering based on the coefficient of variation (< 30% in quality control samples), was conducted using the XCMS package (v3.12.0). In total, 1,090 metabolites were retained for further analysis, comprising 597 positive and 493 negative ion metabolic modes (Supplementary Table S1 ). Hierarchical clustering of the metabolites revealed distinct metabolic signatures between the two groups (Fig. 2 a). The metabolic characteristics of the epilepsy cohorts showed significant distinctions between groups and good consistency within groups, according to the hierarchical clustering dendrogram (Fig. 2 b). The principal component analysis (PCA) results demonstrated that the NDRE group exhibited tighter clustering than the DRE group, with a clear segregation trend between the two groups. The first two principal components (PC1 and PC2) accounted for 51.4% variance (R2X [cum] = 0.514). Notably, the DRE group displayed a greater dispersion along the PC1 axis, indicating that metabolic variations in this direction were closely associated with drug resistance (Fig. 3 a). Plot of partial least squares discriminant analysis (PLS-DA) results showed a more pronounced separation trend between patients with NDRE and DRE [R2X(cum) = 0.233, R2Y(cum) = 0.477, Q2(cum) = -0.12] (Fig. 3 b). Furthermore, the plot of orthogonal PLS-DA (OPLS-DA) results also showed an obvious separation trend between the patients with NDRE and DRE R2X(cum) = 0.233, R2Y(cum) = 0.477, Q2(cum) = -0.0107] (Fig. 3 c). The permutation test results (n = 200) confirmed the robustness of the model fit [Q2(cum)= -0.0107, pQ2(cum) = 0.055], thereby validating the reliability of the subsequent differential metabolite screening process (Fig. 3 d). Differential metabolite identification In all, 1,090 metabolites were found and quantified. Among these, 109 metabolites with significant differences (variable importance in projection [VIP] > 1, P < 0.05) between the groups were screened, including 53 lipid metabolites, 10 organic acids, nine steroid metabolites, seven terpenoid metabolites, two amino acids, one folic acid, and 27 other metabolites (Fig. 4 a, Supplementary Table S2). The volcano plot (Fig. 4 b) illustrates differential metabolites that are upregulated and downregulated based on the fold-change in metabolite levels in the DRE group relative to those in the NDRE group. In order to illustrate the general trend and extent of variation in metabolite quantitative values, a Z-score plot was also created using values derived from the relative content of metabolites (Supplementary Fig. S1 ) [ 14 ]. According to the heat map of differential metabolites (Fig. 5 , Supplementary Table S2), 22 metabolites, such as Isoleucyl-Tryptophan, PE- NMe2(18:3(6Z,9Z,12Z)/20:5(5Z,8Z,11Z,14Z,17Z)), PS (18:3(9Z,12Z,15Z)/ 20:0), 3-hydroxyhexadecanoyl carnitine and 2-amino-5,6-dichloro-3,4-dihydroquinazoline, were up-regulated in the DRE group. Conversely, 87 metabolites, including Desacetyllaurenobiolide, ROSAVIN, and 15(S)-hydroxyeicosatrienoic acid, were downregulated in the DRE group. Furthermore, the distribution features of the 109 differential metabolites between the DRE and NDRE groups were visually represented by the rain cloud plot (Supplementary Fig. S2). Pathway enrichment analysis The metabolites that showed notable alterations between the two groups were subjected to Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis. A total of 16 significantly enriched metabolic pathways were identified, with the top five being choline metabolism in cancer ( P = 0.001), sphingolipid metabolism ( P = 0.005), glycerophospholipid metabolism ( P = 0.012), one-carbon pool by folate ( P = 0.027), and cholesterol metabolism ( P = 0.031) (Fig. 6 a, Supplementary Table S3). Additionally, we analyzed how different metabolites affected the two groups' metabolic pathways in relation to one another. (Fig. 6 b, Supplementary Table S4). Among these, retinol, sphingolipid, porphyrin, and glycerophospholipid metabolism were the most affected (impact > 0.1 and P < 0.05). When comparing the DRE group with the NDRE group, the enrichment analysis network diagram showed that most of the differential metabolites in these metabolic pathways were considerably downregulated (Fig. 6 c). Predictive model construction Through the integrated feature selection process, based on differential metabolites and one clinical characteristic, we screened out a marker composed of LysoPC(22:1(13Z)), PE-NMe2(18:3(6Z,9Z,12Z) /20:5(5Z,8Z,11Z,14Z,17Z)), PC(18:1(9Z) /18:1(9Z)), 8,11-Eicosadiynoic acid, 5,8,11-Dodecat-riynoic acid, 13-Nor-6-eremophilene-8,11-dione, 27-Hydroxyisomangiferolic acid, Cortolone-3-glucu-ronide, Hovenidulcioside A2, and (2-Benzyl-2,4,6-trihydroxy-2,3-dihydro-1-benzofuran-3-yl) oxida-nesulfonic acid and one clinical characteristic (number of seizures before diagnosis) (Supplementary Table S5). Based on the contributions of these variables to the predicted outcomes, we calculated the feature importance scores (Fig. 7 a). Based on these biomarkers, the support vector machine (SVM) machine-learning algorithm was used to construct the classification model. Furthermore, 34 patients with NDRE and seven with DRE served as the test group for evaluating the model's predictive power. The predictive model demonstrated a stable predictive performance, with a cross-validated area under the curve (AUC) of 0.810 and an independent test set AUC of 0.794 (Fig. 7 b, c). In the training cohort, DRE's probability of disease (POD) index was greater than NDRE's, according to the metabolite marker set, and its AUC was 96.85 (Fig. 7 d, e). Additionally, the findings demonstrated that, with an AUC of 79.41, the POD index of DRE was likewise greater than that of NDRE (Fig. 7 f, g). These findings suggest that DRE can be predicted using particular serum metabolites. Discussion Currently, drug resistance is a pressing problem in the treatment of epilepsy since it is associated with high mortality, significant burden, severe psychosocial dysfunction, and a lower quality of life. Developing reliable early warning system that could help clinicians identify people with DRE is therefore critically important. Although numerous predictors and risk factors have been identified, including clinical characteristics and biochemical indicators with some epigenetic and genetic markers, their sensitivity and specificity remain poor. To date, there has been no effective early warning model for DRE. Enzymes and metabolic substrates have become desirable molecular targets for promoting recovery and preventing epileptic seizures [ 13 ]. The ability of small-molecule serum metabolites to penetrate the blood-brain barrier and potentially reflect biological processes taking place in the central nervous system makes them an excellent clinical diagnostic medium. A deeper exploration of the baseline metabolic characteristics and construction of an early warning system for patients with DRE could assist clinicians in developing therapy plans and offer novel ideas for the creation of successful anti-epilepsy strategies. In this prospective cohort study, we comprehensively investigated the baseline serum metabolomic characteristics of patients with epilepsy. Using liquid chromatography-mass spectrometry (LC-MS), 1,090 metabolites were found and quantified using non-targeted metabolic analysis. The DRE and NDRE groups' early serum metabolite profiles were compared, and 109 metabolites—the majority of which were lipids—with notable distinctions between the groups (VIP > 1, P < 0.05) were identified. This is consistent with previous studies that have found that patients with DRE have lipid and amino acid metabolic disorder [ 15 – 17 ]. We further conducted KEGG enrichment analysis and found that it mainly involved pathways including choline metabolism in cancer, sphingolipid metabolism, glycerophospholipid metabolism, one-carbon pool by folate, and cholesterol metabolism. The majority of the differential metabolites in these metabolic pathways were considerably downregulated in the DRE group when compared to the NDRE group, according to our mapping of the differential metabolites to the respective metabolic pathways. In addition, by integrating untargeted metabolomics with machine learning algorithms, we further evaluated the predictive model's effectiveness in the test cohort after identifying a set of 11 metabolites and one clinical feature. The model effectively predicted the early onset of DRE with an AUC of 0.810 in the cross-validation cohort and 0.794 in the independent testing cohort. These metabolic dysregulation phenomena observed in patients with DRE are consistent with the evidence linking metabolic dysfunction to neuronal hyperexcitation and neuroinflammation. We observed that, compared with patients with NDRE, the levels of metabolites involved in choline metabolism in cancer and glycerophospholipid metabolism were decreased in patients with DRE. In the central nervous system, glycerophospholipids are essential for preserving normal functions of synapses, receptors, transport proteins, neurotransmitters, and signal transduction processes [ 18 ]. Previous studies have showed that glycerophospholipid metabolic disorders are present in patients with Alzheimer's disease (AD) and depression [ 19 , 20 ] as well as epilepsy, where the level of glycerophospholipids is positively associated with the decrease in epileptic seizures [ 21 , 22 ]. Abnormal glycerophospholipid metabolism impairs structural remodeling and excitability of neuronal cell membranes and interferes with neuronal signal transduction [ 23 ]. Some studies have shown that dysregulation of phosphatidylcholine catabolism leads to an imbalance in γ-aminobutyric acid-mediated neuronal inhibition, thereby triggering epileptic seizures [ 24 ]. Compared with humans, phosphatidylethanolamine (PE) exhibits distinct changes in animal models of epilepsy. Whether PE levels decrease or increase, these alterations may disrupt the lipid composition balance and regulate epilepsy by influencing lipid-sensitive substances, such as membrane proteins [ 25 ]. Another early metabolic dysregulation observed in patients with DRE is reduced levels of metabolites associated with sphingolipid metabolism. Notably, sphingosine-1-phosphate (S1P), a bioactive sphingolipid metabolite, plays a pivotal role in modulating essential neurophysiological processes, including neuronal development, differentiation, migration, and survival [ 26 ]. S1P can decrease neuronal apoptosis through improving mitochondrial function and structural damage, according to mechanistic studies using both in vivo and in vitro models. This improves the cognitive, emotional, and behavioral problems linked to epilepsy and helps control epileptic seizures [ 27 ]. Folate and one-carbon metabolism disorders have extensive potential effects, including impaired DNA transcription, methylation, and epigenetic modifications. These alterations may influence tissue growth, differentiation, repair processes, and the balance between excitatory and inhibitory mechanisms [ 28 ]. Folate and its derivatives exhibit significant convulsive properties and have been shown to experimentally reverse the anticonvulsant effects of certain drugs. This evidence suggests that folate metabolism may be involved in the antiepileptic effects of drugs and in some potential seizure mechanisms [ 29 ]. Homocysteine (Hcy) levels serve as biomarkers for folate and one-carbon metabolic disorders. Previous studies have demonstrated that new antiepileptic drugs such as levetiracetam, oxcarbazepine, and topiramate are significantly correlated with increased Hcy concentrations [ 30 ]. Patients treated with valproic acid or lamotrigine have notably lower serum folate levels [ 31 ]. Severe developmental disorders and epilepsy result from cerebral folate shortage, which is characterized by decreased levels of 5-methyltetrahydrofolate, an active folate metabolite, in the cerebrospinal fluid while retaining normal levels outside the central nervous system [ 32 ]. However, our study revealed that patients with DRE had decreased levels of metabolites involved in one carbon pool via the folate pathway compared with patients with NDRE. This finding may be related to the dosage of the antiepileptic drugs and other potential mechanisms. Decreased levels of metabolites involved in sphingolipid metabolism are another early metabolic feature of patients with refractory epilepsy. Studies have shown that cholesterol influences epilepsy; developmental and epileptic encephalopathies have garnered increasing attention. Dysregulation of the enzyme cholesterol 24-hydroxylase (CH24H) and its product 24-hydroxycholesterol modulates neuronal excitability through multiple mechanisms [ 33 ]. The highly specific CH24H inhibitor soticlestat reduced epileptic seizure behavior and enhanced survival rates in animal models of epilepsy [ 34 , 35 ]. Additionally, a phase 2 clinical study showed that the median frequency of seizures (for the entire patient group) and the frequency of convulsive seizures (for the Dravet syndrome cohort) demonstrated a statistically significant and clinically meaningful decrease following soticlestat treatment [ 36 ]. Thus, this prospective cohort study analyzed the baseline metabolic characteristics of DRE and provided a non-invasive, serum-based tool for the early identification of the potential population with DRE. These findings enhance our understanding of the metabolic mechanisms underlying DRE, provide new ideas for targeted metabolic therapy, aid clinicians in facilitating early personalized treatment strategies, and ultimately alleviate the burden on patients with epilepsy. There are several drawbacks on this study. First, our findings could not have been as broadly applicable due to the single-center design and rather small sample size. Multicenter studies with larger sample sizes would be more conducive to elucidating the clinical significance of these metabolites. Second, although untargeted metabolomics was employed to describe early metabolite characteristics in patients with DRE and NDRE, absolute quantification of candidate metabolites could not be performed, which lacks the reliability provided by targeted metabolomics analysis. Finally, although we identified several metabolites and metabolic pathways associated with DRE development, more investigation is needed to fully understand the underlying pathophysiological mechanisms. Materials and Methods Research program and participants Between 2022 and 2025, this prospective cohort study was carried out at the Epilepsy Center of the Neurology Department of Zhengzhou University's First Affiliated Hospital. The inclusion criteria for patients with confirmed epilepsy were as follows: 1) one unprovoked (or reflex) seizure and a probability of additional seizures comparable to the general recurrence risk (at least 60%) following two unprovoked seizures over the following ten years; 2) at least two unprovoked (or reflex) seizures that occur more than 24 h apart; and 3) a diagnosis of epilepsy syndrome. The exclusion criteria for patients with epilepsy were as follows: 1) have received antibiotics, probiotics, prebiotics, vitamin, protein, or unsaturated fatty acids within the last three months; 2) a history of autoimmune disorders, including rheumatoid arthritis, systemic lupus erythematosus, multiple sclerosis, and neuromyelitis optica; 3) current or past history of malignancy and gastrointestinal surgery; 4) a history of neurological or psychiatric disorders such as Parkinson's disease, AD, anxiety disorder, depression, autism spectrum disorder, schizophrenia; 5) have other metabolic diseases such as hypertension, diabetes, obesity, and metabolic syndrome; 6) undergoing irregular ASM therapy and poor compliance. A prospective cohort was established in 2021, and 151 patients with epilepsy were recruited. After these participants were recruited, we collected their basic information, clinical characteristics, and serum samples, including clinical data, laboratory and imaging examinations, and EEG. They were interviewed through clinical visits or phone calls every six months for four years. At each follow-up, information on seizure occurrence, medication use, medication adherence and changes, and potential adverse reactions were collected. In addition, patients were required to contact the treating physician within a few days of experiencing a seizure. Based on the DRE diagnostic criteria, patients were split into DRE and NDRE groups [ 2 ]. After propensity score matching of baseline data such as the sex and age, the DRE (n = 33) and NDRE groups (n = 89) were ultimately included and then randomly assigned to the training and test sets at a 2:1 ratio. Serum sample collection Following collection into a serum tube (BD, Oxford, UK), venous blood was centrifuged at 4°C and 10,000 rpm for 10 minutes and stored at -80°C. For analysis, the samples were thawed at 4°C. Subsequently, 400 µL of methanol containing 0.02 mg/mL internal standard (L-2-chlorophenylalanine) was added to a 100 µL aliquot of the sample in a 1.5 mL tube. The mixture was vortexed for 30 seconds and extracted ultrasonically for 30 minutes at 5°C (40 kHz). After incubation at -20°C for 30 minutes, the samples were centrifuged at 4°C and 13,000 g for 15 minutes. The supernatant was then moved to an LC-MS container for examination. A quality control (QC) sample was also made by pooling a 20 µL aliquot of supernatant from each sample. LC–MS experiment A tandem ultra-high-performance liquid chromatography-Fourier transform mass spectrometry system (UHPLC-Q Exactive HF-X, Thermo Fisher Scientific, USA) was used to conduct the LC-MS analysis. The chromatographic column was an ACQUITY UPLC HSS T3 (100 mm × 2.1 mm internal diameter, 1.8 µm; Waters Corporation, Milford, MA, USA). The mobile phases were as follows: A) 5% H2O / 5% ACN (with 0.1% formic acid); B) 47.5% ACN / 47.5% IPA / 5% H2O (with 0.1% formic acid). The sample injection volume was 2 uL, and the column temperature was maintained at 40°C. After electrospraying these materials, mass spectrum data were obtained using both the positive and negative ion scanning modes. Equal quantities of each sample extract were combined to create QC samples, which were then processed in the same way as the analytical samples. Every five to fifteen analytical samples, a QC sample was analyzed to monitor the system's stability. Chromatogram of total ions The mass spectrometer continually received the chromatographically separated sample's constituents, and the mass spectrum was continuously scanned in order to gather data. Each scan produced a mass spectrum, and the total ion current intensity was calculated by adding the ion intensities in each mass spectrum. The ordinate was the sum of the ionic strengths, and the abscissa was the total ion chromatograms for time. The total ion chromatograms of the QC samples in both positive and negative ion modes were used to evaluate the detection performance. Identification of metabolites Raw data for baseline filtering, peak alignment, integration, retention time correction, and peak identification were imported using Progenesis QI (Waters Corporation, Milford, MA, USA). Ultimately, a data matrix was obtained that included data on peak intensity, mass-to-charge ratio, and retention duration. The metabolic database was compared with the MS and MS/MS mass spectrum data. The secondary mass spectrum matching score was used to identify the metabolites, and the MS mass error was set to be less than 10 ppm. The primary metabolite identification databases include popular public and self-built databases including MetaboAnalyst ( https://www.metaboanalyst.ca/ ), the Human Metabolome Database ( https://hmdb.ca/ ) and Metabolite Link ( https://metlin.scripps.edu/ ). Multivariate data analysis The dimensionality reduction of the sample data was achieved by PCA, PLS-DA, and OPLS-DA using the \"Ropls\" package (v1.34.0) of R 4.4.1 ( https://www.r-project.org/ ). Permutation tests were used to overfit the model; Q2 shows the model's predictive power, and R2X and R2Y indicate the model's interpretation rates for the X and Y matrices, respectively. A better model fit and a more accurate classification of the training set samples into their original categories are indicated by values nearer one. The VIP was computed using OPLS-DA dimensionality reduction, and the P -value was ascertained using statistical testing. To help identify marker metabolites, the fold-change was calculated to assess the impact and explanatory capacity of each metabolite's content on sample categorization and discrimination. When the P value was < 0.05 and the VIP value was > 1, metabolites were considered statistically significant. Advanced analysis MetaboAnalystR (v4.0.0) was used to perform KEGG pathway enrichment analysis of the differential metabolite list (KEGG PATHWAY Database, https://www.genome.jp/kegg/pathway.html ) [ 37 ]. The hypergeometric distribution test served as the foundation for the enrichment analysis. The Relative Betweenness Centrality metric was used for the topological study. Construction of predictive models This project utilized the mlr3 framework (mlr3verse v0.3.1) to implement the training and evaluation of the machine learning models. The ensemble feature selection (EFS) method was applied for feature selection, and SVM machine learning algorithms were employed to identify robust metabolic biomarkers. Considering that the DRE and NDRE groups in the training set differed significantly in one clinical parameter (number of seizures before diagnosis) and that previous studies reported that the number of seizures before diagnosis was a predictor of DRE, we screened it through EFS along with metabolites. The data underwent preprocessing steps, including the removal of low-variance features and imputation of missing values, followed by a hierarchical stratified division into the training (2/3) and test (1/3) sets. Hyperparameters were optimized via random search, and the model performance was rigorously assessed using 5-fold cross-validation as well as independent testing. The feature importance was analyzed using algorithm-specific metrics, and repeatability was ensured through parallel computing, fixed random seeds, and modularized pipelines. Receiver operating characteristic (ROC) curves were used to illustrate the results, highlighting the model's key predictive attributes and capacity for generalization. The \"pROC\" package (v1.18.5) was used to construct the ROC curve [ 38 ], and the ggplot2 package (v3.5.1) was used to draw the graphics. Statistical analysis The t-test is used to compare the two groups' continuous variables, which have a normal distribution and are displayed as the mean ± standard deviation. The Wilcoxon rank-sum test was used to compare groups, and continuous variables that were not normally distributed are displayed as medians and interquartile ranges. To compare categorical variables between the two groups, the chi-square or Fisher's exact test was employed. The Wilcox test/t test package (v4.4.0) and SPSS V.27 for Windows (IBM SPSS, Chicago, IL, USA) were used for statistical analyses. The threshold for statistical significance was set at P < 0.05. Conclusions In this prospective cohort study, we analyzed the baseline serum metabolic signatures between patients with DRE and NDRE and developed a non-invasive, serum-based predictive model for identifying the potential population of patients with DRE at an early stage, with a cross-validated AUC of 0.810 and an independent test set AUC of 0.794. These findings enhance our understanding of the metabolic mechanisms underlying DRE, provide new ideas for targeted metabolic therapy, aid clinicians in facilitating early personalized treatment strategies, and ultimately alleviate the burden on patients with epilepsy. Declarations Acknowledgements We express our gratitude to all the kind volunteer participants in the study. Author contributions Conceptualization, ZR and YJ; investigation, WK, QH, MC and LX; data curation, WK, QH, ZZ, RL and YM; writing—original draft preparation, WK, ZR and YJ; visualization, YF; supervision, ZR and YJ; project administration, ZR and YJ; funding acquisition, ZR and YJ. All authors have read and agreed to the published version of the manuscript. Competing interests The authors declare no competing interests. Data availability The serum metabolomics data sets supporting the conclusions of this article are included within the article and its Supplementary Materials. Further inquiries can be directed to the corresponding author. Ethics declarations The Institutional Review Board of Zhengzhou University's First Affiliated Hospital gave its approval to this study (No. 2021-KY-0574-002). The Helsinki Declaration and the Guidelines for Good Clinical Practice were followed when conducting the study. Informed consent was obtained from all participants involved in the study. Funding This study was sponsored by grants from Key project of Henan Provincial Natural Science Foundation (HNSZRKXJJZDXM2023019), Key Scientific Research Project of Henan Province University (24A320031), Henan Province Science and Technology Project (242102311134), Henan University Science and Technology Innovation Team Support Program (25IRTSTHN040). Consent to participate/Consent to publish Not applicable References Banerjee, P.N., D. Filippi, and W. Allen Hauser, The descriptive epidemiology of epilepsy-a review. Epilepsy Res, 2009. 85 (1): p. 31-45. Kwan, P., et al., Definition of drug resistant epilepsy: consensus proposal by the ad hoc Task Force of the ILAE Commission on Therapeutic Strategies. Epilepsia, 2010. 51 (6): p. 1069-77. Kalilani, L., et al., The epidemiology of drug-resistant epilepsy: A systematic review and meta-analysis. Epilepsia, 2018. 59 (12): p. 2179-2193. Kwan, P., S.C. Schachter, and M.J. Brodie, Drug-resistant epilepsy. N Engl J Med, 2011. 365 (10): p. 919-26. Janmohamed, M., M.J. Brodie, and P. Kwan, Pharmacoresistance - Epidemiology, mechanisms, and impact on epilepsy treatment. Neuropharmacology, 2020. 168 : p. 107790. Anyanwu, C. and G.K. Motamedi, Diagnosis and Surgical Treatment of Drug-Resistant Epilepsy. Brain Sci, 2018. 8 (4). Li, Z., et al., Potential clinical and biochemical markers for the prediction of drug-resistant epilepsy: A literature review. Neurobiol Dis, 2022. 174 : p. 105872. Shao, Y. and W. Le, Recent advances and perspectives of metabolomics-based investigations in Parkinson's disease. Mol Neurodegener, 2019. 14 (1): p. 3. Bhargava, P. and D.C. Anthony, Metabolomics in multiple sclerosis disease course and progression. Mult Scler, 2020. 26 (5): p. 591-598. Goodacre, R., et al., Metabolomics by numbers: acquiring and understanding global metabolite data. Trends Biotechnol, 2004. 22 (5): p. 245-52. Scheffer, I.E., et al., ILAE classification of the epilepsies: Position paper of the ILAE Commission for Classification and Terminology. Epilepsia, 2017. 58 (4): p. 512-521. Lin Lin Lee, V., et al., Treatment, Therapy and Management of Metabolic Epilepsy: A Systematic Review. Int J Mol Sci, 2018. 19 (3). Rho, J.M. and D. Boison, The metabolic basis of epilepsy. Nat Rev Neurol, 2022. 18 (6): p. 333-347. Rao, G., J. Sui, and J. Zhang, Metabolomics reveals significant variations in metabolites and correlations regarding the maturation of walnuts (Juglans regia L.). Biol Open, 2016. 5 (6): p. 829-36. Wu, H.C., et al., Altered metabolomic-genomic signature: A potential noninvasive biomarker of epilepsy. Epilepsia, 2017. 58 (9): p. 1626-1636. Detour, J., et al., Metabolomic characterization of human hippocampus from drug-resistant epilepsy with mesial temporal seizure. Epilepsia, 2018. 59 (3): p. 607-616. Zhang, H., et al., Untargeted lipidomic analysis of human hippocampus for temporal lobe epilepsy with hippocampal sclerosis. Epilepsy Res, 2020. 161 : p. 106299. Frisardi, V., et al., Glycerophospholipids and glycerophospholipid-derived lipid mediators: a complex meshwork in Alzheimer's disease pathology. Prog Lipid Res, 2011. 50 (4): p. 313-30. Xu, S., et al., Spatially and temporally probing distinctive glycerophospholipid alterations in Alzheimer's disease mouse brain via high-resolution ion mobility-enabled sn-position resolved lipidomics. Nat Commun, 2024. 15 (1): p. 6252. Xie, J., et al., Multi-omics data reveals the important role of glycerophospholipid metabolism in the crosstalk between gut and brain in depression. J Transl Med, 2023. 21 (1): p. 93. Li, Z., et al., Role of Glycerophospholipid Metabolism in Epilepsy. Curr Neuropharmacol, 2025. Dahlin, M., C.E. Wheelock, and S. Prast-Nielsen, Association between seizure reduction during ketogenic diet treatment of epilepsy and changes in circulatory metabolites and gut microbiota composition. EBioMedicine, 2024. 109 : p. 105400. Chen, Y., et al., Integrated Proteomics and Lipidomics Analysis of Hippocampus to Reveal the Metabolic Landscape of Epilepsy. ACS Omega, 2025. 10 (9): p. 9351-9367. Kim, H.Y., P.G. Suh, and J.I. Kim, The Role of Phospholipase C in GABAergic Inhibition and Its Relevance to Epilepsy. Int J Mol Sci, 2021. 22 (6). Qiu, X., et al., Integrative analysis of non-targeted lipidomic data and brain structural imaging identifies phosphatidylethanolamine associated with epileptogenesis. Metabolomics, 2020. 16 (10): p. 110. van Echten-Deckert, G., The role of sphingosine 1-phosphate metabolism in brain health and disease. Pharmacol Ther, 2023. 244 : p. 108381. Ben, X., et al., Metabolomics-driven exploration of sphingosine 1-phosphate mechanisms in refractory epilepsy. Neurobiol Dis, 2025. 212 : p. 106953. Reynolds, E.H., Antiepileptic drugs, folate and one carbon metabolism revisited. Epilepsy Behav, 2020. 112 : p. 107336. Baxter, M.G., A.A. Miller, and R.A. Webster, Some studies on the convulsant action of folic acid. Br J Pharmacol, 1973. 48 (2): p. 350p-351p. Kim, D.W., et al., Effects of new antiepileptic drugs on circulatory markers for vascular risk in patients with newly diagnosed epilepsy. Epilepsia, 2013. 54 (10): p. e146-9. Ni, G., et al., Effects of antiepileptic drug monotherapy on one-carbon metabolism and DNA methylation in patients with epilepsy. PLoS One, 2015. 10 (4): p. e0125656. Ramaekers, V.T. and E.V. Quadros, Cerebral Folate Deficiency Syndrome: Early Diagnosis, Intervention and Treatment Strategies. Nutrients, 2022. 14 (15). Wheless, J.W. and J.M. Rho, Role of cholesterol in modulating brain hyperexcitability. Epilepsia, 2025. 66 (1): p. 33-46. Hawkins, N.A., et al., Soticlestat, a novel cholesterol 24-hydroxylase inhibitor, reduces seizures and premature death in Dravet syndrome mice. Epilepsia, 2021. 62 (11): p. 2845-2857. Nishi, T., et al., Anticonvulsive properties of soticlestat, a novel cholesterol 24-hydroxylase inhibitor. Epilepsia, 2022. 63 (6): p. 1580-1590. Hahn, C.D., et al., A phase 2, randomized, double-blind, placebo-controlled study to evaluate the efficacy and safety of soticlestat as adjunctive therapy in pediatric patients with Dravet syndrome or Lennox-Gastaut syndrome (ELEKTRA). Epilepsia, 2022. 63 (10): p. 2671-2683. Xia, J. and D.S. Wishart, Web-based inference of biological patterns, functions and pathways from metabolomic data using MetaboAnalyst. Nat Protoc, 2011. 6 (6): p. 743-60. Robin, X., et al., pROC: an open-source package for R and S+ to analyze and compare ROC curves. BMC Bioinformatics, 2011. 12 : p. 77. Additional Declarations No competing interests reported. Supplementary Files Supplementarymaterials.zip Cite Share Download PDF Status: Published Journal Publication published 03 Dec, 2025 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Revision requested 19 Sep, 2025 Reviews received at journal 13 Sep, 2025 Reviewers agreed at journal 03 Sep, 2025 Reviews received at journal 27 Aug, 2025 Reviewers agreed at journal 25 Aug, 2025 Reviewers invited by journal 11 Aug, 2025 Editor invited by journal 23 Jul, 2025 Editor assigned by journal 23 Jul, 2025 Submission checks completed at journal 22 Jul, 2025 First submitted to journal 21 Jul, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {\"props\":{\"pageProps\":{\"initialData\":{\"identity\":\"rs-7173411\",\"acceptedTermsAndConditions\":true,\"allowDirectSubmit\":false,\"archivedVersions\":[],\"articleType\":\"Article\",\"associatedPublications\":[],\"authors\":[{\"id\":500380471,\"identity\":\"967dd901-9eb5-4a33-856d-a69b105afe1e\",\"order_by\":0,\"name\":\"Wenzhong Kang\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"The First Affiliated Hospital of Zhengzhou University\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Wenzhong\",\"middleName\":\"\",\"lastName\":\"Kang\",\"suffix\":\"\"},{\"id\":500380472,\"identity\":\"55b1e016-3ea6-42c7-b999-9d152b40e385\",\"order_by\":1,\"name\":\"Qingrong Han\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"The First Affiliated Hospital of Zhengzhou University\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Qingrong\",\"middleName\":\"\",\"lastName\":\"Han\",\"suffix\":\"\"},{\"id\":500380473,\"identity\":\"166e414d-e85f-4171-8914-87f564c3977c\",\"order_by\":2,\"name\":\"Min Chen\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"The First Affiliated Hospital of Zhengzhou University\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Min\",\"middleName\":\"\",\"lastName\":\"Chen\",\"suffix\":\"\"},{\"id\":500380474,\"identity\":\"4866d6f9-0f58-4d72-a333-625fd95a9d75\",\"order_by\":3,\"name\":\"Longxing Xue\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"The First Affiliated Hospital of Zhengzhou University\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Longxing\",\"middleName\":\"\",\"lastName\":\"Xue\",\"suffix\":\"\"},{\"id\":500380475,\"identity\":\"37623de5-a522-4c0f-acb6-70a6f54fc655\",\"order_by\":4,\"name\":\"Zhihua Zhao\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"The First Affiliated Hospital of Zhengzhou University\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Zhihua\",\"middleName\":\"\",\"lastName\":\"Zhao\",\"suffix\":\"\"},{\"id\":500380476,\"identity\":\"f640e2ed-ff19-41c0-ab3c-010259e04830\",\"order_by\":5,\"name\":\"Ruihan Liu\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"Central South University\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Ruihan\",\"middleName\":\"\",\"lastName\":\"Liu\",\"suffix\":\"\"},{\"id\":500380477,\"identity\":\"cf9f4efc-9dde-4462-b2e5-42cc1fd6a551\",\"order_by\":6,\"name\":\"Yarong Ma\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"The First Affiliated Hospital of Zhengzhou University\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Yarong\",\"middleName\":\"\",\"lastName\":\"Ma\",\"suffix\":\"\"},{\"id\":500380478,\"identity\":\"a4cc9cdd-de38-40f2-b15b-ed640e9c615f\",\"order_by\":7,\"name\":\"Yanqiu Fu\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"The First Affiliated Hospital of Zhengzhou University\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Yanqiu\",\"middleName\":\"\",\"lastName\":\"Fu\",\"suffix\":\"\"},{\"id\":500380479,\"identity\":\"77745acc-267f-4ba9-9bb0-b58cb859b1b6\",\"order_by\":8,\"name\":\"Zhigang Ren\",\"email\":\"\",\"orcid\":\"\",\"institution\":\"The First Affiliated Hospital of Zhengzhou University\",\"correspondingAuthor\":false,\"prefix\":\"\",\"firstName\":\"Zhigang\",\"middleName\":\"\",\"lastName\":\"Ren\",\"suffix\":\"\"},{\"id\":500380480,\"identity\":\"de60208b-3d5c-497d-8e19-a864193ab669\",\"order_by\":9,\"name\":\"Yan Jiang\",\"email\":\"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA4ElEQVRIiWNgGAWjYBACxmYQaWDDw8/MfPBBQkUNsVoK0uQk29uSDR6cOUasXR8OGxv0nFGTfNjCTFgxczvzs4dfDJgTN0jksFUkNrAx8Ld3JxBwGJu5sYwBW+J2idxjNxJ3yDBInDm7gZBfzKQlDHgSd87IS7uReIaNwUAil5AW9m9ALRKJG27kmBUktjETo4XHTPKDgYGxwZkzZgzEaimTZjBIAAeyRMKZYzwE/WLYf3yb5I8//8FR+fFHRY0cf3svAS0NwIDmQRLgwaUSDuRBjvtBUNkoGAWjYBSMaAAAnxhIjPaQblUAAAAASUVORK5CYII=\",\"orcid\":\"\",\"institution\":\"The First Affiliated Hospital of Zhengzhou University\",\"correspondingAuthor\":true,\"prefix\":\"\",\"firstName\":\"Yan\",\"middleName\":\"\",\"lastName\":\"Jiang\",\"suffix\":\"\"}],\"badges\":[],\"createdAt\":\"2025-07-21 05:23:05\",\"currentVersionCode\":1,\"declarations\":\"\",\"doi\":\"10.21203/rs.3.rs-7173411/v1\",\"doiUrl\":\"https://doi.org/10.21203/rs.3.rs-7173411/v1\",\"draftVersion\":[],\"editorialEvents\":[{\"content\":\"https://doi.org/10.1038/s41598-025-30741-8\",\"type\":\"published\",\"date\":\"2025-12-03T15:58:04+00:00\"}],\"editorialNote\":\"\",\"failedWorkflow\":false,\"files\":[{\"id\":89402444,\"identity\":\"0eaf2642-26aa-4e77-bb90-97ee5ed497a5\",\"added_by\":\"auto\",\"created_at\":\"2025-08-19 14:32:27\",\"extension\":\"png\",\"order_by\":1,\"title\":\"Figure 1\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":1126870,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eFlowchart of the study. After strict inclusion and exclusion criteria, 151 patients with epilepsy are prospectively recruited. Details of basic information, clinical characteristics, and serum samples of the patients are collected. They are interviewed through clinical visits or phone calls every six months for four years. According to the diagnostic criteria of DRE, the patients are divided into the DRE and NDRE groups. After propensity score matching of baseline data (sex and age), the DRE (n = 33) and NDRE groups (n = 89) are ultimately included, and they are then randomly assigned to the training and test sets at a 2:1 ratio. Identification of differential metabolites, pathway enrichment analysis, and predictive model construction are analyzed in the training set. This model is further validated using the test set. DRE, drug-resistant epilepsy; NDRE, non-drug-refractory epilepsy.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"Figure1.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7173411/v1/bb9df028dc58c632b9286350.png\"},{\"id\":89402004,\"identity\":\"529b08d2-06fe-44f4-afe6-27f3805fa8b7\",\"added_by\":\"auto\",\"created_at\":\"2025-08-19 14:24:27\",\"extension\":\"png\",\"order_by\":2,\"title\":\"Figure 2\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":1653088,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eSerum metabolomic profiles and multivariate analysis of metabolic separation between the NDRE and DRE groups. (\\u003cstrong\\u003ea\\u003c/strong\\u003e) Each metabolite's relative expression across all samples is displayed in the heat map of the overall metabolite cluster. The expression is higher when the color is redder. The expression is lower when the color is bluer. (\\u003cstrong\\u003eb\\u003c/strong\\u003e) Hierarchical clustering dendrogram of epilepsy cohorts. Closer and shorter cluster trees represent higher similarity between samples. DRE, drug-resistant epilepsy; NDRE, non-drug-refractory epilepsy.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"Figure2.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7173411/v1/09f1a46ba5eb2ca4e44a5739.png\"},{\"id\":89402023,\"identity\":\"ee20c414-349c-42b6-a00f-65370c838c3b\",\"added_by\":\"auto\",\"created_at\":\"2025-08-19 14:24:28\",\"extension\":\"png\",\"order_by\":3,\"title\":\"Figure 3\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":803181,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eMultivariate analysis of metabolic separation between the NDRE and DRE groups.\\u003cstrong\\u003e \\u003c/strong\\u003e(\\u003cstrong\\u003ea\\u003c/strong\\u003e) PCA score plot showing dispersion of drug-resistant epilepsy (DRE) and non-resistant (NDRE) groups along PC1. (\\u003cstrong\\u003eb\\u003c/strong\\u003e) Plot of PLS-DA results shows a more pronounced separation trend between patients with NDRE and DRE. (\\u003cstrong\\u003ec\\u003c/strong\\u003e) Plot of OPLS-DA results also shows an obvious separation trend between patients with NDRE and DRE. (\\u003cstrong\\u003ed\\u003c/strong\\u003e) The permutation test results (n = 200) show the robustness of the model fit. DRE, drug-resistant epilepsy; NDRE, non-drug-refractory epilepsy; PCA, principal component analysis; PLS-DA, partial least squares discriminant analysis; OPLS-DA, orthogonal partial least squares discriminant analysis.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"Figure3.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7173411/v1/daed15609ff469b894db805a.png\"},{\"id\":89402451,\"identity\":\"00ce0928-060f-46d9-a572-ed9d2b41a1bd\",\"added_by\":\"auto\",\"created_at\":\"2025-08-19 14:32:28\",\"extension\":\"png\",\"order_by\":4,\"title\":\"Figure 4\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":294812,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eIdentification of differentially abundant metabolites. (\\u003cstrong\\u003ea\\u003c/strong\\u003e) Histogram graph of the number of differential metabolites shows 109 metabolites with significant differences between groups. (\\u003cstrong\\u003eb\\u003c/strong\\u003e) The volcano plot visually illustrates the distribution and change trend of upregulated and downregulated metabolites in the two groups. DRE, drug-resistant epilepsy; NDRE, non-drug-refractory epilepsy; VIP, variable importance in projection.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"Figure4.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7173411/v1/5861d0478f2c7420ff132b8d.png\"},{\"id\":89402453,\"identity\":\"bd3d074b-5859-4d14-9b6b-fbad6eb498f3\",\"added_by\":\"auto\",\"created_at\":\"2025-08-19 14:32:28\",\"extension\":\"png\",\"order_by\":5,\"title\":\"Figure 5\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":3417574,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003eHierarchical clustering of differentially abundant metabolites between the NDRE and DRE groups. Samples are represented by columns, metabolites by rows, and the sample cluster tree is at the top, followed by the different metabolite cluster tree on the left. The gradient color represents the quantitative value's magnitude; the higher the expression, the redder the color, and the lower the expression, the bluer the color. DRE, drug-resistant epilepsy; NDRE, non-drug-refractory epilepsy.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"Figure5.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7173411/v1/de37b48210c21508b56cd51f.png\"},{\"id\":89402029,\"identity\":\"367c5d58-1ba9-40c5-b6b1-c988fe68f637\",\"added_by\":\"auto\",\"created_at\":\"2025-08-19 14:24:28\",\"extension\":\"png\",\"order_by\":6,\"title\":\"Figure 6\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":1146886,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003ePathway enrichment analysis of different metabolites between the DRE and NDRE groups. (\\u003cstrong\\u003ea\\u003c/strong\\u003e) The enriched metabolic pathways of differential metabolites are displayed in the enrichment analysis histogram. (\\u003cstrong\\u003eb\\u003c/strong\\u003e) The relative impact of differential metabolites on metabolic pathways in the two groups is displayed in a KEGG topological bubble diagram. (\\u003cstrong\\u003ec\\u003c/strong\\u003e) A visual representation of the regulatory network of metabolic pathways and differential metabolites is provided by the KEGG pathway network diagram. KEGG, Kyoto Encyclopedia of Genes and Genomes; DRE, drug-resistant epilepsy; NDRE, non-drug-refractory epilepsy.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"Figure6.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7173411/v1/b58556c81a9de97c8eba9805.png\"},{\"id\":89402025,\"identity\":\"203ebeb3-be18-4965-aaff-d42b24e7e1e2\",\"added_by\":\"auto\",\"created_at\":\"2025-08-19 14:24:28\",\"extension\":\"png\",\"order_by\":7,\"title\":\"Figure 7\",\"display\":\"\",\"copyAsset\":false,\"role\":\"figure\",\"size\":1265385,\"visible\":true,\"origin\":\"\",\"legend\":\"\\u003cp\\u003ePredictive model construction and assessment. (\\u003cstrong\\u003ea\\u003c/strong\\u003e) Variable importance in the model. (\\u003cstrong\\u003eb\\u003c/strong\\u003e) The ROC curve has a cross-validated AUC of 0.810 (95% CI: 0.704–0.915) between DRE and NDRE in the training cohort. (\\u003cstrong\\u003ec\\u003c/strong\\u003e) The ROC curve has a cross-validated AUC of 0.794 (95% CI: 0.626–0.962) between DRE and NDRE in the test cohort. (\\u003cstrong\\u003ed\\u003c/strong\\u003e) The POD index of DRE is greater than that of NDRE in the training cohort. (\\u003cstrong\\u003ee\\u003c/strong\\u003e) The POD index has an AUC of 96.85 (95% CI: 93.83–99.88, \\u003cem\\u003eP\\u003c/em\\u003e \\u0026lt; 0.0001). (\\u003cstrong\\u003ef\\u003c/strong\\u003e) The POD index of DRE is greater than that of NDRE in the test cohort. (\\u003cstrong\\u003eg\\u003c/strong\\u003e) The POD index has an AUC of 79.41 (95% CI: 62.65–96.17, \\u003cem\\u003eP\\u003c/em\\u003e= 0.0136). ROC, receiver operating characteristic; AUC, area under curve; POD, probability of disease; DRE, drug-resistant epilepsy; NDRE, non-drug-refractory epilepsy.\\u003c/p\\u003e\",\"description\":\"\",\"filename\":\"Figure7.png\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7173411/v1/4d9c55607975174a74bb6e0e.png\"},{\"id\":97724631,\"identity\":\"2596ea2b-301a-4c94-8174-24e67f3b50bb\",\"added_by\":\"auto\",\"created_at\":\"2025-12-08 16:13:00\",\"extension\":\"pdf\",\"order_by\":0,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"manuscript-pdf\",\"size\":11646410,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"manuscript.pdf\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7173411/v1/b353ef9f-f5a5-497e-88c2-b92e029a9ced.pdf\"},{\"id\":89402452,\"identity\":\"a682d6e1-4f19-4ac1-9f6c-48bfae229ebc\",\"added_by\":\"auto\",\"created_at\":\"2025-08-19 14:32:28\",\"extension\":\"zip\",\"order_by\":1,\"title\":\"\",\"display\":\"\",\"copyAsset\":false,\"role\":\"supplement\",\"size\":6889923,\"visible\":true,\"origin\":\"\",\"legend\":\"\",\"description\":\"\",\"filename\":\"Supplementarymaterials.zip\",\"url\":\"https://assets-eu.researchsquare.com/files/rs-7173411/v1/6adbc8858a6c0f26e96592c7.zip\"}],\"financialInterests\":\"No competing interests reported.\",\"formattedTitle\":\"Alterations in serum metabolomics predict drug-resistant epilepsy\",\"fulltext\":[{\"header\":\"Introduction\",\"content\":\"\\u003cp\\u003eEpilepsy is a prevalent chronic neurological disorder distinguished by the occurrence of unprovoked seizures and has a global prevalence of 0.5\\u0026ndash;1% in developed countries and significantly higher rates in developing countries [\\u003cspan citationid=\\\"CR1\\\" class=\\\"CitationRef\\\"\\u003e1\\u003c/span\\u003e]. Although medication can control seizures in most patients, some patients still experience uncontrollable seizures due to drug resistance. In 2010, the International League Against Epilepsy (ILAE) defined drug-resistant epilepsy (DRE) as the failure of two appropriately chosen and tolerated anti-seizure medication (ASM) regimens (monotherapy or polytherapy) to achieve freedom from seizure for more than one year [\\u003cspan citationid=\\\"CR2\\\" class=\\\"CitationRef\\\"\\u003e2\\u003c/span\\u003e]. Systematic reviews have estimated the prevalence at 30%, with an annual incidence of 15% [\\u003cspan citationid=\\\"CR3\\\" class=\\\"CitationRef\\\"\\u003e3\\u003c/span\\u003e]. Patients with DRE have a higher risk of premature mortality, neuropsychological impairment, and a significant reduction in quality of life [\\u003cspan citationid=\\\"CR4\\\" class=\\\"CitationRef\\\"\\u003e4\\u003c/span\\u003e]. Consequently, it is crucial to identify DRE in patients as soon as possible.\\u003c/p\\u003e\\u003cp\\u003eThe mechanisms behind drug resistance in epilepsy are still unknown. Current diagnostic paradigms remain reactive and rely on the outcome of medical treatment rather than predictive biomarkers, which hinders early individualization and optimization of treatment for patients diagnosed with DRE [\\u003cspan citationid=\\\"CR5\\\" class=\\\"CitationRef\\\"\\u003e5\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR6\\\" class=\\\"CitationRef\\\"\\u003e6\\u003c/span\\u003e]. Several models and predictors have been proposed to predict DRE, including clinical and biochemical markers such as age at onset, abnormal electroencephalogram (EEG), seizure type, initial frequency of seizures, family history of epilepsy, aberrant neuroimaging, levels of high mobility group box 1, and \\u003cem\\u003eSCN1A\\u003c/em\\u003e polymorphisms [\\u003cspan citationid=\\\"CR7\\\" class=\\\"CitationRef\\\"\\u003e7\\u003c/span\\u003e]. However, there are currently no molecular biomarkers able to identify possible DRE populations early on with good sensitivity and specificity.\\u003c/p\\u003e\\u003cp\\u003eThe measurement of small molecule concentrations in blood, urine, cerebrospinal fluid, and other biological matrices is known as metabolomics [\\u003cspan citationid=\\\"CR8\\\" class=\\\"CitationRef\\\"\\u003e8\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR9\\\" class=\\\"CitationRef\\\"\\u003e9\\u003c/span\\u003e]. It offers data for ensuing \\\"omics\\\" technologies like proteomics and genomics. Metabolomics has become a significant method for the identification of disease biomarkers, characterized by its efficiency and high-throughput precision. It not only unravels intricate biological mechanisms but also serves as a reliable way to monitor therapeutic effects. Metabolomics is widely recognized as an effective approach to disease diagnosis, classification, staging, treatment planning, and prognostic evaluation [\\u003cspan citationid=\\\"CR10\\\" class=\\\"CitationRef\\\"\\u003e10\\u003c/span\\u003e]. In 2017, the ILAE classified \\u0026ldquo;metabolic etiology\\u0026rdquo; as one of six major etiological categories of epilepsy [\\u003cspan citationid=\\\"CR11\\\" class=\\\"CitationRef\\\"\\u003e11\\u003c/span\\u003e]. There are several subtypes of metabolic epilepsy, such as mitochondrial diseases, folinic acid-responsive seizures, and cerebral folate insufficiency [\\u003cspan citationid=\\\"CR12\\\" class=\\\"CitationRef\\\"\\u003e12\\u003c/span\\u003e]. Basic research and translational studies have shown that the relationship between cerebral metabolism and seizures is complicated and bi-directional and forms a harmful cycle that exacerbates the negative effects of seizures. Metabolic moleculars and enzymes have turned into appealing biological targets for preventing epileptic seizures and fostering recuperation, and metabolism-based therapy methods, such as high-fat antiepileptic ketogenic diets, have progressively gained traction [\\u003cspan citationid=\\\"CR13\\\" class=\\\"CitationRef\\\"\\u003e13\\u003c/span\\u003e].\\u003c/p\\u003e\\u003cp\\u003eTherefore, this study aimed to elucidate the differences in baseline serum metabolomics between DRE and non-drug-refractory epilepsy (NDRE) through a prospective cohort study. This study aimed to clarify the material basis for the occurrence and progression of DRE from a metabolic perspective, thereby providing a biological foundation and novel insights for further investigation of the pathophysiological mechanisms underlying DRE. Additionally, a predictive model was developed utilizing the clinical traits and potential metabolites identified in the training cohort and subsequently confirmed in the test cohort.\\u003c/p\\u003e\"},{\"header\":\"Results\",\"content\":\"\\u003cp\\u003e\\u003cb\\u003eParticipants\\u0026rsquo; clinical features and basic information\\u003c/b\\u003e\\u003c/p\\u003e\\u003cp\\u003eThe study flow chart is displayed in Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig1\\\" class=\\\"InternalRef\\\"\\u003e1\\u003c/span\\u003e. The clinical characteristics and demographic information of the study population are presented in Table\\u0026nbsp;\\u003cspan refid=\\\"Tab1\\\" class=\\\"InternalRef\\\"\\u003e1\\u003c/span\\u003e. Other than the number of seizures prior to diagnosis, which also did not exhibit a statistically significant difference (\\u003cem\\u003eP\\u003c/em\\u003e\\u0026thinsp;\\u0026gt;\\u0026thinsp;0.05), there were no statistically significant differences between the DRE and NDRE groups (\\u003cem\\u003eP\\u003c/em\\u003e\\u0026thinsp;\\u0026gt;\\u0026thinsp;0.05). In the training cohort, the DRE group exhibited more frequent seizures before diagnosis than the NDRE group (\\u003cem\\u003eP\\u003c/em\\u003e\\u0026thinsp;=\\u0026thinsp;0.004). Additionally, the NDRE group exhibited greater levels of total cholesterol than the DRE group within the test cohort (3.97 vs. 3.44 mmol/L, \\u003cem\\u003eP\\u003c/em\\u003e\\u0026thinsp;=\\u0026thinsp;0.011), but they were still within the normal physiological range.\\u003c/p\\u003e\\u003cp\\u003e\\u003c/p\\u003e\\u003cp\\u003e\\u003cdiv class=\\\"gridtable\\\"\\u003e\\u003ctable float=\\\"Yes\\\" id=\\\"Tab1\\\" border=\\\"1\\\"\\u003e\\u003ccaption language=\\\"En\\\"\\u003e\\u003cdiv class=\\\"CaptionNumber\\\"\\u003eTable 1\\u003c/div\\u003e\\u003cdiv class=\\\"CaptionContent\\\"\\u003e\\u003cp\\u003eThe study cohort's demographic data and clinical features.\\u003c/p\\u003e\\u003c/div\\u003e\\u003c/caption\\u003e\\u003ccolgroup cols=\\\"7\\\"\\u003e\\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c1\\\" colnum=\\\"1\\\"\\u003e\\u003c/div\\u003e\\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c2\\\" colnum=\\\"2\\\"\\u003e\\u003c/div\\u003e\\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c3\\\" colnum=\\\"3\\\"\\u003e\\u003c/div\\u003e\\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c4\\\" colnum=\\\"4\\\"\\u003e\\u003c/div\\u003e\\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c5\\\" colnum=\\\"5\\\"\\u003e\\u003c/div\\u003e\\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c6\\\" colnum=\\\"6\\\"\\u003e\\u003c/div\\u003e\\u003cdiv align=\\\"left\\\" class=\\\"colspec\\\" colname=\\\"c7\\\" colnum=\\\"7\\\"\\u003e\\u003c/div\\u003e\\u003cthead\\u003e\\u003ctr\\u003e\\u003cth align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eItem\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colspan=\\\"3\\\" nameend=\\\"c4\\\" namest=\\\"c2\\\"\\u003e\\u003cp\\u003eTrain cohort\\u003c/p\\u003e\\u003c/th\\u003e\\u003cth align=\\\"left\\\" colspan=\\\"3\\\" nameend=\\\"c7\\\" namest=\\\"c5\\\"\\u003e\\u003cp\\u003eTest cohort\\u003c/p\\u003e\\u003c/th\\u003e\\u003c/tr\\u003e\\u003c/thead\\u003e\\u003ctbody\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003eDRE (n\\u0026thinsp;=\\u0026thinsp;26)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003eNDRE (n\\u0026thinsp;=\\u0026thinsp;55)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e\\u003cem\\u003eP\\u003c/em\\u003e- value\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003eDRE (n\\u0026thinsp;=\\u0026thinsp;7)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003eNDRE (n\\u0026thinsp;=\\u0026thinsp;34)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e\\u003cp\\u003e\\u003cem\\u003eP\\u003c/em\\u003e- value\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eFemale sex, n (%)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e11 (42.31)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e21 (38.18)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e0.723\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e4 (57.14)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e17 (50.00)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e\\u003cp\\u003e1.000\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eWeight (kg), mean (SD)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e58.2 (17.8)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e57.7 (17.3)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e0.905\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e59.2 (18.2)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e59.2 (18.2)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e\\u003cp\\u003e0.077\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eAge (years), n (%)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e0.570\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e\\u003cp\\u003e0.910\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eChildhood (0\\u0026ndash;10)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e4 (15.38)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e7 (12.73)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e1 (14.29)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e7 (20.59)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eAdolescent (11\\u0026ndash;17)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e6 (23.08)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e19 (34.55)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e2 (28.57)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e8 (23.53)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eAdult age (\\u0026ge;\\u0026thinsp;18)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e16 (61.54)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e29 (52.73)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e4 (57.14)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e19 (55.88)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eAge at onset of epilepsy (years), n (%)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e0.874\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e\\u003cp\\u003e0.057\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eChildhood (0\\u0026ndash;10)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e8(30.77)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e14(25.45)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e4(57.14)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e10(29.41)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eAdolescent (11\\u0026ndash;17)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e8(30.77)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e19(34.55)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e3(42.86)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e12(35.29)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eAdult age (\\u0026ge;\\u0026thinsp;18)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e10(38.46)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e22(40.00)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e0(0.00)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e12(35.29)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eNumber of seizures before diagnosis, n (%)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e0.004\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e\\u003cp\\u003e0.194\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e\\u0026lt;4\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e10 (38.46)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e39 (70.91)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e4 (57.14)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e21 (61.76)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e4\\u0026ndash;10\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e2 (7.69)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e8 (14.55)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e0 (0.00)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e6 (17.65)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e10\\u0026ndash;100\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e12 (46.15)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e7 (12.73)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e3 (42.86)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e7 (20.59)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003e\\u0026gt;\\u0026thinsp;100\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e2 (7.69)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e1 (1.82)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e0 (0.00)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e0 (0.00)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eMRI, n (%)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e0.310\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e\\u003cp\\u003e0.060\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eAbnormal\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e14 (53.85)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e23 (41.82)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e6 (57.14)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e13 (38.24)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eNormal\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e12 (46.15)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e32 (58.18)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e1 (42.86)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e21 (61.76)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eElectroencephalogram, n (%)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e0.053\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e\\u003cp\\u003e0.615\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eAbnormal\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e20 (76.92)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e30 (54.55)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e4 (57.14)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e13 (38.24)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eNormal\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e6 (23.08)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e25 (45.45)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e3 (42.86)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e21 (61.76)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e\\u0026nbsp;\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eWBC (x10*9), mean (SD)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e5.70 (1.38)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e6.04 (1.67)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e0.382\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e5.92 (1.14)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e5.44 (1.50)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e\\u003cp\\u003e0.434\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eRBC (x10*9), mean (SD)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e4.37 (0.46)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e4.27(0.58)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e0.448\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e4.04 (0.58)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e4.28 (0.52)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e\\u003cp\\u003e0.287\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eHb (g/L), mean (SD)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e131 (14)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e131 (14)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e0.911\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e135 (14)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e132 (14)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e\\u003cp\\u003e0.629\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003ePLT (x10*6), mean (SD)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e222 (82)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e232 (75)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e0.593\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e209 (42)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e231 (71)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e\\u003cp\\u003e0.448\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eALB (g/L), mean (SD)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e44.3 (5.0)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e43.2 (5.9)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e0.424\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e42.0 (2.3)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e43.3 (6.6)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e\\u003cp\\u003e0.615\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eTCHO (mmol/L), mean (SD)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e3.82 (0.40)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e3.79 (0.43)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e0.720\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e3.44 (0.39)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e3.97 (0.49)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e\\u003cp\\u003e0.011\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eTG (mmol/L), median (IQR)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e0.99 (0.64,1.37)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e0.72 (0.56,1.28)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e0.230\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e1.15 (0.74,1.33)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e0.74 (0.59,1.24)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e\\u003cp\\u003e0.100\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003ctr\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c1\\\"\\u003e\\u003cp\\u003eTBIL (umol/L) median (IQR)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c2\\\"\\u003e\\u003cp\\u003e6.3 (4.4,8.5)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c3\\\"\\u003e\\u003cp\\u003e6.2 (4.7,8.3)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c4\\\"\\u003e\\u003cp\\u003e0.686\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c5\\\"\\u003e\\u003cp\\u003e5.3 (5.1,6.3)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c6\\\"\\u003e\\u003cp\\u003e6.0 (4.5,7.4)\\u003c/p\\u003e\\u003c/td\\u003e\\u003ctd align=\\\"left\\\" colname=\\\"c7\\\"\\u003e\\u003cp\\u003e0.615\\u003c/p\\u003e\\u003c/td\\u003e\\u003c/tr\\u003e\\u003c/tbody\\u003e\\u003c/colgroup\\u003e\\u003c/table\\u003e\\u003c/div\\u003e\\u003c/p\\u003e\\u003cp\\u003eWhile continuous variables are displayed as mean (standard deviation, SD) or median (interquartile range, IQR), categorical variables are displayed as percentages [n, (%)]. MRI, magnetic resonance imaging; WBC, white blood cells; RBC, red blood cells; Hb, hemoglobin; PLT, platelet; ALB, albumin; TCHO, total cholesterol; TG, triglyceride; TBIL, total bilirubin; DRE, drug-resistant epilepsy; NDRE, non-drug-refractory epilepsy.\\u003c/p\\u003e\\u003cp\\u003e\\u003cb\\u003eMetabolomic profiling and data quality assessment\\u003c/b\\u003e\\u003c/p\\u003e\\u003cp\\u003eNon-targeted metabolomic analysis was performed on the serum samples of 81 patients with epilepsy in the training set (55 patients with NDRE and 26 with DRE). Data pre-processing, including peak alignment and filtering based on the coefficient of variation (\\u0026lt;\\u0026thinsp;30% in quality control samples), was conducted using the XCMS package (v3.12.0). In total, 1,090 metabolites were retained for further analysis, comprising 597 positive and 493 negative ion metabolic modes (Supplementary Table \\u003cspan refid=\\\"MOESM1\\\" class=\\\"InternalRef\\\"\\u003eS1\\u003c/span\\u003e). Hierarchical clustering of the metabolites revealed distinct metabolic signatures between the two groups (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig2\\\" class=\\\"InternalRef\\\"\\u003e2\\u003c/span\\u003ea). The metabolic characteristics of the epilepsy cohorts showed significant distinctions between groups and good consistency within groups, according to the hierarchical clustering dendrogram (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig2\\\" class=\\\"InternalRef\\\"\\u003e2\\u003c/span\\u003eb). The principal component analysis (PCA) results demonstrated that the NDRE group exhibited tighter clustering than the DRE group, with a clear segregation trend between the two groups. The first two principal components (PC1 and PC2) accounted for 51.4% variance (R2X [cum]\\u0026thinsp;=\\u0026thinsp;0.514). Notably, the DRE group displayed a greater dispersion along the PC1 axis, indicating that metabolic variations in this direction were closely associated with drug resistance (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig3\\\" class=\\\"InternalRef\\\"\\u003e3\\u003c/span\\u003ea). Plot of partial least squares discriminant analysis (PLS-DA) results showed a more pronounced separation trend between patients with NDRE and DRE [R2X(cum)\\u0026thinsp;=\\u0026thinsp;0.233, R2Y(cum)\\u0026thinsp;=\\u0026thinsp;0.477, Q2(cum) = -0.12] (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig3\\\" class=\\\"InternalRef\\\"\\u003e3\\u003c/span\\u003eb). Furthermore, the plot of orthogonal PLS-DA (OPLS-DA) results also showed an obvious separation trend between the patients with NDRE and DRE R2X(cum)\\u0026thinsp;=\\u0026thinsp;0.233, R2Y(cum)\\u0026thinsp;=\\u0026thinsp;0.477, Q2(cum) = -0.0107] (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig3\\\" class=\\\"InternalRef\\\"\\u003e3\\u003c/span\\u003ec). The permutation test results (n\\u0026thinsp;=\\u0026thinsp;200) confirmed the robustness of the model fit [Q2(cum)= -0.0107, pQ2(cum)\\u0026thinsp;=\\u0026thinsp;0.055], thereby validating the reliability of the subsequent differential metabolite screening process (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig3\\\" class=\\\"InternalRef\\\"\\u003e3\\u003c/span\\u003ed).\\u003c/p\\u003e\\u003cp\\u003e\\u003c/p\\u003e\\u003cp\\u003e\\u003c/p\\u003e\\u003cp\\u003e\\u003cb\\u003eDifferential metabolite identification\\u003c/b\\u003e\\u003c/p\\u003e\\u003cp\\u003eIn all, 1,090 metabolites were found and quantified. Among these, 109 metabolites with significant differences (variable importance in projection [VIP]\\u0026thinsp;\\u0026gt;\\u0026thinsp;1, \\u003cem\\u003eP\\u003c/em\\u003e\\u0026thinsp;\\u0026lt;\\u0026thinsp;0.05) between the groups were screened, including 53 lipid metabolites, 10 organic acids, nine steroid metabolites, seven terpenoid metabolites, two amino acids, one folic acid, and 27 other metabolites (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig4\\\" class=\\\"InternalRef\\\"\\u003e4\\u003c/span\\u003ea, Supplementary Table S2). The volcano plot (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig4\\\" class=\\\"InternalRef\\\"\\u003e4\\u003c/span\\u003eb) illustrates differential metabolites that are upregulated and downregulated based on the fold-change in metabolite levels in the DRE group relative to those in the NDRE group. In order to illustrate the general trend and extent of variation in metabolite quantitative values, a Z-score plot was also created using values derived from the relative content of metabolites (Supplementary Fig. \\u003cspan refid=\\\"MOESM1\\\" class=\\\"InternalRef\\\"\\u003eS1\\u003c/span\\u003e) [\\u003cspan citationid=\\\"CR14\\\" class=\\\"CitationRef\\\"\\u003e14\\u003c/span\\u003e]. According to the heat map of differential metabolites (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig5\\\" class=\\\"InternalRef\\\"\\u003e5\\u003c/span\\u003e, Supplementary Table S2), 22 metabolites, such as Isoleucyl-Tryptophan, PE- NMe2(18:3(6Z,9Z,12Z)/20:5(5Z,8Z,11Z,14Z,17Z)), PS (18:3(9Z,12Z,15Z)/ 20:0), 3-hydroxyhexadecanoyl carnitine and 2-amino-5,6-dichloro-3,4-dihydroquinazoline, were up-regulated in the DRE group. Conversely, 87 metabolites, including Desacetyllaurenobiolide, ROSAVIN, and 15(S)-hydroxyeicosatrienoic acid, were downregulated in the DRE group. Furthermore, the distribution features of the 109 differential metabolites between the DRE and NDRE groups were visually represented by the rain cloud plot (Supplementary Fig. S2).\\u003c/p\\u003e\\u003cp\\u003e\\u003c/p\\u003e\\u003cp\\u003e\\u003c/p\\u003e\\u003cp\\u003e\\u003cb\\u003ePathway enrichment analysis\\u003c/b\\u003e\\u003c/p\\u003e\\u003cp\\u003eThe metabolites that showed notable alterations between the two groups were subjected to Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis. A total of 16 significantly enriched metabolic pathways were identified, with the top five being choline metabolism in cancer (\\u003cem\\u003eP\\u003c/em\\u003e\\u0026thinsp;=\\u0026thinsp;0.001), sphingolipid metabolism (\\u003cem\\u003eP\\u003c/em\\u003e\\u0026thinsp;=\\u0026thinsp;0.005), glycerophospholipid metabolism (\\u003cem\\u003eP\\u003c/em\\u003e\\u0026thinsp;=\\u0026thinsp;0.012), one-carbon pool by folate (\\u003cem\\u003eP\\u003c/em\\u003e\\u0026thinsp;=\\u0026thinsp;0.027), and cholesterol metabolism (\\u003cem\\u003eP\\u003c/em\\u003e\\u0026thinsp;=\\u0026thinsp;0.031) (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig6\\\" class=\\\"InternalRef\\\"\\u003e6\\u003c/span\\u003ea, Supplementary Table S3). Additionally, we analyzed how different metabolites affected the two groups' metabolic pathways in relation to one another. (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig6\\\" class=\\\"InternalRef\\\"\\u003e6\\u003c/span\\u003eb, Supplementary Table S4). Among these, retinol, sphingolipid, porphyrin, and glycerophospholipid metabolism were the most affected (impact\\u0026thinsp;\\u0026gt;\\u0026thinsp;0.1 and \\u003cem\\u003eP\\u003c/em\\u003e\\u0026thinsp;\\u0026lt;\\u0026thinsp;0.05). When comparing the DRE group with the NDRE group, the enrichment analysis network diagram showed that most of the differential metabolites in these metabolic pathways were considerably downregulated (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig6\\\" class=\\\"InternalRef\\\"\\u003e6\\u003c/span\\u003ec).\\u003c/p\\u003e\\u003cp\\u003e\\u003c/p\\u003e\\u003cp\\u003e\\u003cb\\u003ePredictive model construction\\u003c/b\\u003e\\u003c/p\\u003e\\u003cp\\u003eThrough the integrated feature selection process, based on differential metabolites and one clinical characteristic, we screened out a marker composed of LysoPC(22:1(13Z)), PE-NMe2(18:3(6Z,9Z,12Z) /20:5(5Z,8Z,11Z,14Z,17Z)), PC(18:1(9Z) /18:1(9Z)), 8,11-Eicosadiynoic acid, 5,8,11-Dodecat-riynoic acid, 13-Nor-6-eremophilene-8,11-dione, 27-Hydroxyisomangiferolic acid, Cortolone-3-glucu-ronide, Hovenidulcioside A2, and (2-Benzyl-2,4,6-trihydroxy-2,3-dihydro-1-benzofuran-3-yl) oxida-nesulfonic acid and one clinical characteristic (number of seizures before diagnosis) (Supplementary Table S5). Based on the contributions of these variables to the predicted outcomes, we calculated the feature importance scores (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig7\\\" class=\\\"InternalRef\\\"\\u003e7\\u003c/span\\u003ea). Based on these biomarkers, the support vector machine (SVM) machine-learning algorithm was used to construct the classification model. Furthermore, 34 patients with NDRE and seven with DRE served as the test group for evaluating the model's predictive power. The predictive model demonstrated a stable predictive performance, with a cross-validated area under the curve (AUC) of 0.810 and an independent test set AUC of 0.794 (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig7\\\" class=\\\"InternalRef\\\"\\u003e7\\u003c/span\\u003eb, c). In the training cohort, DRE's probability of disease (POD) index was greater than NDRE's, according to the metabolite marker set, and its AUC was 96.85 (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig7\\\" class=\\\"InternalRef\\\"\\u003e7\\u003c/span\\u003ed, e). Additionally, the findings demonstrated that, with an AUC of 79.41, the POD index of DRE was likewise greater than that of NDRE (Fig.\\u0026nbsp;\\u003cspan refid=\\\"Fig7\\\" class=\\\"InternalRef\\\"\\u003e7\\u003c/span\\u003ef, g). These findings suggest that DRE can be predicted using particular serum metabolites.\\u003c/p\\u003e\\u003cp\\u003e\\u003c/p\\u003e\"},{\"header\":\"Discussion\",\"content\":\"\\u003cp\\u003eCurrently, drug resistance is a pressing problem in the treatment of epilepsy since it is associated with high mortality, significant burden, severe psychosocial dysfunction, and a lower quality of life. Developing reliable early warning system that could help clinicians identify people with DRE is therefore critically important. Although numerous predictors and risk factors have been identified, including clinical characteristics and biochemical indicators with some epigenetic and genetic markers, their sensitivity and specificity remain poor. To date, there has been no effective early warning model for DRE. Enzymes and metabolic substrates have become desirable molecular targets for promoting recovery and preventing epileptic seizures [\\u003cspan citationid=\\\"CR13\\\" class=\\\"CitationRef\\\"\\u003e13\\u003c/span\\u003e]. The ability of small-molecule serum metabolites to penetrate the blood-brain barrier and potentially reflect biological processes taking place in the central nervous system makes them an excellent clinical diagnostic medium. A deeper exploration of the baseline metabolic characteristics and construction of an early warning system for patients with DRE could assist clinicians in developing therapy plans and offer novel ideas for the creation of successful anti-epilepsy strategies.\\u003c/p\\u003e\\u003cp\\u003eIn this prospective cohort study, we comprehensively investigated the baseline serum metabolomic characteristics of patients with epilepsy. Using liquid chromatography-mass spectrometry (LC-MS), 1,090 metabolites were found and quantified using non-targeted metabolic analysis. The DRE and NDRE groups' early serum metabolite profiles were compared, and 109 metabolites\\u0026mdash;the majority of which were lipids\\u0026mdash;with notable distinctions between the groups (VIP\\u0026thinsp;\\u0026gt;\\u0026thinsp;1, \\u003cem\\u003eP\\u003c/em\\u003e\\u0026thinsp;\\u0026lt;\\u0026thinsp;0.05) were identified. This is consistent with previous studies that have found that patients with DRE have lipid and amino acid metabolic disorder [\\u003cspan additionalcitationids=\\\"CR16\\\" citationid=\\\"CR15\\\" class=\\\"CitationRef\\\"\\u003e15\\u003c/span\\u003e\\u0026ndash;\\u003cspan citationid=\\\"CR17\\\" class=\\\"CitationRef\\\"\\u003e17\\u003c/span\\u003e]. We further conducted KEGG enrichment analysis and found that it mainly involved pathways including choline metabolism in cancer, sphingolipid metabolism, glycerophospholipid metabolism, one-carbon pool by folate, and cholesterol metabolism. The majority of the differential metabolites in these metabolic pathways were considerably downregulated in the DRE group when compared to the NDRE group, according to our mapping of the differential metabolites to the respective metabolic pathways. In addition, by integrating untargeted metabolomics with machine learning algorithms, we further evaluated the predictive model's effectiveness in the test cohort after identifying a set of 11 metabolites and one clinical feature. The model effectively predicted the early onset of DRE with an AUC of 0.810 in the cross-validation cohort and 0.794 in the independent testing cohort. These metabolic dysregulation phenomena observed in patients with DRE are consistent with the evidence linking metabolic dysfunction to neuronal hyperexcitation and neuroinflammation.\\u003c/p\\u003e\\u003cp\\u003eWe observed that, compared with patients with NDRE, the levels of metabolites involved in choline metabolism in cancer and glycerophospholipid metabolism were decreased in patients with DRE. In the central nervous system, glycerophospholipids are essential for preserving normal functions of synapses, receptors, transport proteins, neurotransmitters, and signal transduction processes [\\u003cspan citationid=\\\"CR18\\\" class=\\\"CitationRef\\\"\\u003e18\\u003c/span\\u003e]. Previous studies have showed that glycerophospholipid metabolic disorders are present in patients with Alzheimer's disease (AD) and depression [\\u003cspan citationid=\\\"CR19\\\" class=\\\"CitationRef\\\"\\u003e19\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR20\\\" class=\\\"CitationRef\\\"\\u003e20\\u003c/span\\u003e] as well as epilepsy, where the level of glycerophospholipids is positively associated with the decrease in epileptic seizures [\\u003cspan citationid=\\\"CR21\\\" class=\\\"CitationRef\\\"\\u003e21\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR22\\\" class=\\\"CitationRef\\\"\\u003e22\\u003c/span\\u003e]. Abnormal glycerophospholipid metabolism impairs structural remodeling and excitability of neuronal cell membranes and interferes with neuronal signal transduction [\\u003cspan citationid=\\\"CR23\\\" class=\\\"CitationRef\\\"\\u003e23\\u003c/span\\u003e]. Some studies have shown that dysregulation of phosphatidylcholine catabolism leads to an imbalance in γ-aminobutyric acid-mediated neuronal inhibition, thereby triggering epileptic seizures [\\u003cspan citationid=\\\"CR24\\\" class=\\\"CitationRef\\\"\\u003e24\\u003c/span\\u003e]. Compared with humans, phosphatidylethanolamine (PE) exhibits distinct changes in animal models of epilepsy. Whether PE levels decrease or increase, these alterations may disrupt the lipid composition balance and regulate epilepsy by influencing lipid-sensitive substances, such as membrane proteins [\\u003cspan citationid=\\\"CR25\\\" class=\\\"CitationRef\\\"\\u003e25\\u003c/span\\u003e].\\u003c/p\\u003e\\u003cp\\u003eAnother early metabolic dysregulation observed in patients with DRE is reduced levels of metabolites associated with sphingolipid metabolism. Notably, sphingosine-1-phosphate (S1P), a bioactive sphingolipid metabolite, plays a pivotal role in modulating essential neurophysiological processes, including neuronal development, differentiation, migration, and survival [\\u003cspan citationid=\\\"CR26\\\" class=\\\"CitationRef\\\"\\u003e26\\u003c/span\\u003e]. S1P can decrease neuronal apoptosis through improving mitochondrial function and structural damage, according to mechanistic studies using both in vivo and in vitro models. This improves the cognitive, emotional, and behavioral problems linked to epilepsy and helps control epileptic seizures [\\u003cspan citationid=\\\"CR27\\\" class=\\\"CitationRef\\\"\\u003e27\\u003c/span\\u003e].\\u003c/p\\u003e\\u003cp\\u003eFolate and one-carbon metabolism disorders have extensive potential effects, including impaired DNA transcription, methylation, and epigenetic modifications. These alterations may influence tissue growth, differentiation, repair processes, and the balance between excitatory and inhibitory mechanisms [\\u003cspan citationid=\\\"CR28\\\" class=\\\"CitationRef\\\"\\u003e28\\u003c/span\\u003e]. Folate and its derivatives exhibit significant convulsive properties and have been shown to experimentally reverse the anticonvulsant effects of certain drugs. This evidence suggests that folate metabolism may be involved in the antiepileptic effects of drugs and in some potential seizure mechanisms [\\u003cspan citationid=\\\"CR29\\\" class=\\\"CitationRef\\\"\\u003e29\\u003c/span\\u003e]. Homocysteine (Hcy) levels serve as biomarkers for folate and one-carbon metabolic disorders. Previous studies have demonstrated that new antiepileptic drugs such as levetiracetam, oxcarbazepine, and topiramate are significantly correlated with increased Hcy concentrations [\\u003cspan citationid=\\\"CR30\\\" class=\\\"CitationRef\\\"\\u003e30\\u003c/span\\u003e]. Patients treated with valproic acid or lamotrigine have notably lower serum folate levels [\\u003cspan citationid=\\\"CR31\\\" class=\\\"CitationRef\\\"\\u003e31\\u003c/span\\u003e]. Severe developmental disorders and epilepsy result from cerebral folate shortage, which is characterized by decreased levels of 5-methyltetrahydrofolate, an active folate metabolite, in the cerebrospinal fluid while retaining normal levels outside the central nervous system [\\u003cspan citationid=\\\"CR32\\\" class=\\\"CitationRef\\\"\\u003e32\\u003c/span\\u003e]. However, our study revealed that patients with DRE had decreased levels of metabolites involved in one carbon pool via the folate pathway compared with patients with NDRE. This finding may be related to the dosage of the antiepileptic drugs and other potential mechanisms.\\u003c/p\\u003e\\u003cp\\u003eDecreased levels of metabolites involved in sphingolipid metabolism are another early metabolic feature of patients with refractory epilepsy. Studies have shown that cholesterol influences epilepsy; developmental and epileptic encephalopathies have garnered increasing attention. Dysregulation of the enzyme cholesterol 24-hydroxylase (CH24H) and its product 24-hydroxycholesterol modulates neuronal excitability through multiple mechanisms [\\u003cspan citationid=\\\"CR33\\\" class=\\\"CitationRef\\\"\\u003e33\\u003c/span\\u003e]. The highly specific CH24H inhibitor soticlestat reduced epileptic seizure behavior and enhanced survival rates in animal models of epilepsy [\\u003cspan citationid=\\\"CR34\\\" class=\\\"CitationRef\\\"\\u003e34\\u003c/span\\u003e, \\u003cspan citationid=\\\"CR35\\\" class=\\\"CitationRef\\\"\\u003e35\\u003c/span\\u003e]. Additionally, a phase 2 clinical study showed that the median frequency of seizures (for the entire patient group) and the frequency of convulsive seizures (for the Dravet syndrome cohort) demonstrated a statistically significant and clinically meaningful decrease following soticlestat treatment [\\u003cspan citationid=\\\"CR36\\\" class=\\\"CitationRef\\\"\\u003e36\\u003c/span\\u003e].\\u003c/p\\u003e\\u003cp\\u003eThus, this prospective cohort study analyzed the baseline metabolic characteristics of DRE and provided a non-invasive, serum-based tool for the early identification of the potential population with DRE. These findings enhance our understanding of the metabolic mechanisms underlying DRE, provide new ideas for targeted metabolic therapy, aid clinicians in facilitating early personalized treatment strategies, and ultimately alleviate the burden on patients with epilepsy.\\u003c/p\\u003e\\u003cp\\u003eThere are several drawbacks on this study. First, our findings could not have been as broadly applicable due to the single-center design and rather small sample size. Multicenter studies with larger sample sizes would be more conducive to elucidating the clinical significance of these metabolites. Second, although untargeted metabolomics was employed to describe early metabolite characteristics in patients with DRE and NDRE, absolute quantification of candidate metabolites could not be performed, which lacks the reliability provided by targeted metabolomics analysis. Finally, although we identified several metabolites and metabolic pathways associated with DRE development, more investigation is needed to fully understand the underlying pathophysiological mechanisms.\\u003c/p\\u003e\"},{\"header\":\"Materials and Methods\",\"content\":\"\\u003cp\\u003e\\u003cb\\u003eResearch program and participants\\u003c/b\\u003e\\u003c/p\\u003e\\u003cp\\u003eBetween 2022 and 2025, this prospective cohort study was carried out at the Epilepsy Center of the Neurology Department of Zhengzhou University's First Affiliated Hospital. The inclusion criteria for patients with confirmed epilepsy were as follows: 1) one unprovoked (or reflex) seizure and a probability of additional seizures comparable to the general recurrence risk (at least 60%) following two unprovoked seizures over the following ten years; 2) at least two unprovoked (or reflex) seizures that occur more than 24 h apart; and 3) a diagnosis of epilepsy syndrome. The exclusion criteria for patients with epilepsy were as follows: 1) have received antibiotics, probiotics, prebiotics, vitamin, protein, or unsaturated fatty acids within the last three months; 2) a history of autoimmune disorders, including rheumatoid arthritis, systemic lupus erythematosus, multiple sclerosis, and neuromyelitis optica; 3) current or past history of malignancy and gastrointestinal surgery; 4) a history of neurological or psychiatric disorders such as Parkinson's disease, AD, anxiety disorder, depression, autism spectrum disorder, schizophrenia; 5) have other metabolic diseases such as hypertension, diabetes, obesity, and metabolic syndrome; 6) undergoing irregular ASM therapy and poor compliance.\\u003c/p\\u003e\\u003cp\\u003eA prospective cohort was established in 2021, and 151 patients with epilepsy were recruited. After these participants were recruited, we collected their basic information, clinical characteristics, and serum samples, including clinical data, laboratory and imaging examinations, and EEG. They were interviewed through clinical visits or phone calls every six months for four years. At each follow-up, information on seizure occurrence, medication use, medication adherence and changes, and potential adverse reactions were collected. In addition, patients were required to contact the treating physician within a few days of experiencing a seizure. Based on the DRE diagnostic criteria, patients were split into DRE and NDRE groups [\\u003cspan citationid=\\\"CR2\\\" class=\\\"CitationRef\\\"\\u003e2\\u003c/span\\u003e]. After propensity score matching of baseline data such as the sex and age, the DRE (n\\u0026thinsp;=\\u0026thinsp;33) and NDRE groups (n\\u0026thinsp;=\\u0026thinsp;89) were ultimately included and then randomly assigned to the training and test sets at a 2:1 ratio.\\u003c/p\\u003e\\u003cp\\u003e\\u003cb\\u003eSerum sample collection\\u003c/b\\u003e\\u003c/p\\u003e\\u003cp\\u003eFollowing collection into a serum tube (BD, Oxford, UK), venous blood was centrifuged at 4\\u0026deg;C and 10,000 rpm for 10 minutes and stored at -80\\u0026deg;C. For analysis, the samples were thawed at 4\\u0026deg;C. Subsequently, 400 \\u0026micro;L of methanol containing 0.02 mg/mL internal standard (L-2-chlorophenylalanine) was added to a 100 \\u0026micro;L aliquot of the sample in a 1.5 mL tube. The mixture was vortexed for 30 seconds and extracted ultrasonically for 30 minutes at 5\\u0026deg;C (40 kHz). After incubation at -20\\u0026deg;C for 30 minutes, the samples were centrifuged at 4\\u0026deg;C and 13,000 g for 15 minutes. The supernatant was then moved to an LC-MS container for examination. A quality control (QC) sample was also made by pooling a 20 \\u0026micro;L aliquot of supernatant from each sample.\\u003c/p\\u003e\\u003cp\\u003e\\u003cb\\u003eLC\\u0026ndash;MS experiment\\u003c/b\\u003e\\u003c/p\\u003e\\u003cp\\u003eA tandem ultra-high-performance liquid chromatography-Fourier transform mass spectrometry system (UHPLC-Q Exactive HF-X, Thermo Fisher Scientific, USA) was used to conduct the LC-MS analysis. The chromatographic column was an ACQUITY UPLC HSS T3 (100 mm \\u0026times; 2.1 mm internal diameter, 1.8 \\u0026micro;m; Waters Corporation, Milford, MA, USA). The mobile phases were as follows: A) 5% H2O / 5% ACN (with 0.1% formic acid); B) 47.5% ACN / 47.5% IPA / 5% H2O (with 0.1% formic acid). The sample injection volume was 2 uL, and the column temperature was maintained at 40\\u0026deg;C. After electrospraying these materials, mass spectrum data were obtained using both the positive and negative ion scanning modes. Equal quantities of each sample extract were combined to create QC samples, which were then processed in the same way as the analytical samples. Every five to fifteen analytical samples, a QC sample was analyzed to monitor the system's stability.\\u003c/p\\u003e\\u003cp\\u003e\\u003cb\\u003eChromatogram of total ions\\u003c/b\\u003e\\u003c/p\\u003e\\u003cp\\u003eThe mass spectrometer continually received the chromatographically separated sample's constituents, and the mass spectrum was continuously scanned in order to gather data. Each scan produced a mass spectrum, and the total ion current intensity was calculated by adding the ion intensities in each mass spectrum. The ordinate was the sum of the ionic strengths, and the abscissa was the total ion chromatograms for time. The total ion chromatograms of the QC samples in both positive and negative ion modes were used to evaluate the detection performance.\\u003c/p\\u003e\\u003cp\\u003e\\u003cb\\u003eIdentification of metabolites\\u003c/b\\u003e\\u003c/p\\u003e\\u003cp\\u003eRaw data for baseline filtering, peak alignment, integration, retention time correction, and peak identification were imported using Progenesis QI (Waters Corporation, Milford, MA, USA). Ultimately, a data matrix was obtained that included data on peak intensity, mass-to-charge ratio, and retention duration. The metabolic database was compared with the MS and MS/MS mass spectrum data. The secondary mass spectrum matching score was used to identify the metabolites, and the MS mass error was set to be less than 10 ppm. The primary metabolite identification databases include popular public and self-built databases including MetaboAnalyst (\\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://www.metaboanalyst.ca/\\u003c/span\\u003e\\u003cspan address=\\\"https://www.metaboanalyst.ca/\\\" targettype=\\\"URL\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e), the Human Metabolome Database (\\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://hmdb.ca/\\u003c/span\\u003e\\u003cspan address=\\\"https://hmdb.ca/\\\" targettype=\\\"URL\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e) and Metabolite Link (\\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://metlin.scripps.edu/\\u003c/span\\u003e\\u003cspan address=\\\"https://metlin.scripps.edu/\\\" targettype=\\\"URL\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e).\\u003c/p\\u003e\\u003cp\\u003e\\u003cb\\u003eMultivariate data analysis\\u003c/b\\u003e\\u003c/p\\u003e\\u003cp\\u003eThe dimensionality reduction of the sample data was achieved by PCA, PLS-DA, and OPLS-DA using the \\\"Ropls\\\" package (v1.34.0) of R 4.4.1 (\\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://www.r-project.org/\\u003c/span\\u003e\\u003cspan address=\\\"https://www.r-project.org/\\\" targettype=\\\"URL\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e). Permutation tests were used to overfit the model; Q2 shows the model's predictive power, and R2X and R2Y indicate the model's interpretation rates for the X and Y matrices, respectively. A better model fit and a more accurate classification of the training set samples into their original categories are indicated by values nearer one. The VIP was computed using OPLS-DA dimensionality reduction, and the \\u003cem\\u003eP\\u003c/em\\u003e-value was ascertained using statistical testing. To help identify marker metabolites, the fold-change was calculated to assess the impact and explanatory capacity of each metabolite's content on sample categorization and discrimination. When the \\u003cem\\u003eP\\u003c/em\\u003e value was \\u0026lt;\\u0026thinsp;0.05 and the VIP value was \\u0026gt;\\u0026thinsp;1, metabolites were considered statistically significant.\\u003c/p\\u003e\\u003cp\\u003e\\u003cb\\u003eAdvanced analysis\\u003c/b\\u003e\\u003c/p\\u003e\\u003cp\\u003eMetaboAnalystR (v4.0.0) was used to perform KEGG pathway enrichment analysis of the differential metabolite list (KEGG PATHWAY Database, \\u003cspan class=\\\"ExternalRef\\\"\\u003e\\u003cspan class=\\\"RefSource\\\"\\u003ehttps://www.genome.jp/kegg/pathway.html\\u003c/span\\u003e\\u003cspan address=\\\"https://www.genome.jp/kegg/pathway.html\\\" targettype=\\\"URL\\\" class=\\\"RefTarget\\\"\\u003e\\u003c/span\\u003e\\u003c/span\\u003e) [\\u003cspan citationid=\\\"CR37\\\" class=\\\"CitationRef\\\"\\u003e37\\u003c/span\\u003e]. The hypergeometric distribution test served as the foundation for the enrichment analysis. The Relative Betweenness Centrality metric was used for the topological study.\\u003c/p\\u003e\\u003cp\\u003e\\u003cb\\u003eConstruction of predictive models\\u003c/b\\u003e\\u003c/p\\u003e\\u003cp\\u003eThis project utilized the mlr3 framework (mlr3verse v0.3.1) to implement the training and evaluation of the machine learning models. The ensemble feature selection (EFS) method was applied for feature selection, and SVM machine learning algorithms were employed to identify robust metabolic biomarkers. Considering that the DRE and NDRE groups in the training set differed significantly in one clinical parameter (number of seizures before diagnosis) and that previous studies reported that the number of seizures before diagnosis was a predictor of DRE, we screened it through EFS along with metabolites. The data underwent preprocessing steps, including the removal of low-variance features and imputation of missing values, followed by a hierarchical stratified division into the training (2/3) and test (1/3) sets. Hyperparameters were optimized via random search, and the model performance was rigorously assessed using 5-fold cross-validation as well as independent testing. The feature importance was analyzed using algorithm-specific metrics, and repeatability was ensured through parallel computing, fixed random seeds, and modularized pipelines. Receiver operating characteristic (ROC) curves were used to illustrate the results, highlighting the model's key predictive attributes and capacity for generalization. The \\\"pROC\\\" package (v1.18.5) was used to construct the ROC curve [\\u003cspan citationid=\\\"CR38\\\" class=\\\"CitationRef\\\"\\u003e38\\u003c/span\\u003e], and the ggplot2 package (v3.5.1) was used to draw the graphics.\\u003c/p\\u003e\\u003cdiv id=\\\"Sec5\\\" class=\\\"Section2\\\"\\u003e\\u003ch2\\u003eStatistical analysis\\u003c/h2\\u003e\\u003cp\\u003eThe t-test is used to compare the two groups' continuous variables, which have a normal distribution and are displayed as the mean\\u0026thinsp;\\u0026plusmn;\\u0026thinsp;standard deviation. The Wilcoxon rank-sum test was used to compare groups, and continuous variables that were not normally distributed are displayed as medians and interquartile ranges. To compare categorical variables between the two groups, the chi-square or Fisher's exact test was employed. The Wilcox test/t test package (v4.4.0) and SPSS V.27 for Windows (IBM SPSS, Chicago, IL, USA) were used for statistical analyses. The threshold for statistical significance was set at \\u003cem\\u003eP\\u003c/em\\u003e\\u0026thinsp;\\u0026lt;\\u0026thinsp;0.05.\\u003c/p\\u003e\\u003c/div\\u003e\"},{\"header\":\"Conclusions\",\"content\":\"\\u003cp\\u003eIn this prospective cohort study, we analyzed the baseline serum metabolic signatures between patients with DRE and NDRE and developed a non-invasive, serum-based predictive model for identifying the potential population of patients with DRE at an early stage, with a cross-validated AUC of 0.810 and an independent test set AUC of 0.794. These findings enhance our understanding of the metabolic mechanisms underlying DRE, provide new ideas for targeted metabolic therapy, aid clinicians in facilitating early personalized treatment strategies, and ultimately alleviate the burden on patients with epilepsy.\\u003c/p\\u003e\"},{\"header\":\"Declarations\",\"content\":\"\\u003cp\\u003e\\u003cstrong\\u003eAcknowledgements\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eWe express our gratitude to all the kind volunteer participants in the study.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eAuthor contributions\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eConceptualization, ZR and YJ; investigation, WK, QH, MC and LX; data curation, WK, QH, ZZ, RL and YM; writing—original draft preparation, WK, ZR and YJ; visualization, YF; supervision, ZR and YJ; project administration, ZR and YJ; funding acquisition, ZR and YJ. All authors have read and agreed to the published version of the manuscript.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eCompeting interests\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eThe authors declare no competing interests.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eData availability\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eThe serum metabolomics data sets supporting the conclusions of this article are included within the article and its Supplementary Materials. Further inquiries can be directed to the corresponding author.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eEthics declarations\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eThe Institutional Review Board of Zhengzhou University's First Affiliated Hospital gave its approval to this study (No. 2021-KY-0574-002). The Helsinki Declaration and the Guidelines for Good Clinical Practice were followed when conducting the study. Informed consent was obtained from all participants involved in the study.\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eFunding\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eThis study was sponsored by grants from\\u0026nbsp;Key project of Henan Provincial Natural Science Foundation (HNSZRKXJJZDXM2023019), Key Scientific Research Project of Henan Province University (24A320031), Henan Province Science and Technology Project (242102311134), Henan University Science and Technology Innovation Team Support Program (25IRTSTHN040).\\u003c/p\\u003e\\n\\u003cp\\u003e\\u003cstrong\\u003eConsent to participate/Consent to publish\\u003c/strong\\u003e\\u003c/p\\u003e\\n\\u003cp\\u003eNot applicable\\u003c/p\\u003e\"},{\"header\":\"References\",\"content\":\"\\u003col\\u003e\\n\\u003cli\\u003eBanerjee, P.N., D. Filippi, and W. Allen Hauser, \\u003cem\\u003eThe descriptive epidemiology of epilepsy-a review.\\u003c/em\\u003e Epilepsy Res, 2009. \\u003cstrong\\u003e85\\u003c/strong\\u003e(1): p. 31-45.\\u003c/li\\u003e\\n\\u003cli\\u003eKwan, P., et al., \\u003cem\\u003eDefinition of drug resistant epilepsy: consensus proposal by the ad hoc Task Force of the ILAE Commission on Therapeutic Strategies.\\u003c/em\\u003e Epilepsia, 2010. \\u003cstrong\\u003e51\\u003c/strong\\u003e(6): p. 1069-77.\\u003c/li\\u003e\\n\\u003cli\\u003eKalilani, L., et al., \\u003cem\\u003eThe epidemiology of drug-resistant epilepsy: A systematic review and meta-analysis.\\u003c/em\\u003e Epilepsia, 2018. \\u003cstrong\\u003e59\\u003c/strong\\u003e(12): p. 2179-2193.\\u003c/li\\u003e\\n\\u003cli\\u003eKwan, P., S.C. Schachter, and M.J. Brodie, \\u003cem\\u003eDrug-resistant epilepsy.\\u003c/em\\u003e N Engl J Med, 2011. \\u003cstrong\\u003e365\\u003c/strong\\u003e(10): p. 919-26.\\u003c/li\\u003e\\n\\u003cli\\u003eJanmohamed, M., M.J. Brodie, and P. Kwan, \\u003cem\\u003ePharmacoresistance - Epidemiology, mechanisms, and impact on epilepsy treatment.\\u003c/em\\u003e Neuropharmacology, 2020. \\u003cstrong\\u003e168\\u003c/strong\\u003e: p. 107790.\\u003c/li\\u003e\\n\\u003cli\\u003eAnyanwu, C. and G.K. Motamedi, \\u003cem\\u003eDiagnosis and Surgical Treatment of Drug-Resistant Epilepsy.\\u003c/em\\u003e Brain Sci, 2018. \\u003cstrong\\u003e8\\u003c/strong\\u003e(4).\\u003c/li\\u003e\\n\\u003cli\\u003eLi, Z., et al., \\u003cem\\u003ePotential clinical and biochemical markers for the prediction of drug-resistant epilepsy: A literature review.\\u003c/em\\u003e Neurobiol Dis, 2022. \\u003cstrong\\u003e174\\u003c/strong\\u003e: p. 105872.\\u003c/li\\u003e\\n\\u003cli\\u003eShao, Y. and W. Le, \\u003cem\\u003eRecent advances and perspectives of metabolomics-based investigations in Parkinson\\u0026apos;s disease.\\u003c/em\\u003e Mol Neurodegener, 2019. \\u003cstrong\\u003e14\\u003c/strong\\u003e(1): p. 3.\\u003c/li\\u003e\\n\\u003cli\\u003eBhargava, P. and D.C. Anthony, \\u003cem\\u003eMetabolomics in multiple sclerosis disease course and progression.\\u003c/em\\u003e Mult Scler, 2020. \\u003cstrong\\u003e26\\u003c/strong\\u003e(5): p. 591-598.\\u003c/li\\u003e\\n\\u003cli\\u003eGoodacre, R., et al., \\u003cem\\u003eMetabolomics by numbers: acquiring and understanding global metabolite data.\\u003c/em\\u003e Trends Biotechnol, 2004. \\u003cstrong\\u003e22\\u003c/strong\\u003e(5): p. 245-52.\\u003c/li\\u003e\\n\\u003cli\\u003eScheffer, I.E., et al., \\u003cem\\u003eILAE classification of the epilepsies: Position paper of the ILAE Commission for Classification and Terminology.\\u003c/em\\u003e Epilepsia, 2017. \\u003cstrong\\u003e58\\u003c/strong\\u003e(4): p. 512-521.\\u003c/li\\u003e\\n\\u003cli\\u003eLin Lin Lee, V., et al., \\u003cem\\u003eTreatment, Therapy and Management of Metabolic Epilepsy: A Systematic Review.\\u003c/em\\u003e Int J Mol Sci, 2018. \\u003cstrong\\u003e19\\u003c/strong\\u003e(3).\\u003c/li\\u003e\\n\\u003cli\\u003eRho, J.M. and D. Boison, \\u003cem\\u003eThe metabolic basis of epilepsy.\\u003c/em\\u003e Nat Rev Neurol, 2022. \\u003cstrong\\u003e18\\u003c/strong\\u003e(6): p. 333-347.\\u003c/li\\u003e\\n\\u003cli\\u003eRao, G., J. Sui, and J. Zhang, \\u003cem\\u003eMetabolomics reveals significant variations in metabolites and correlations regarding the maturation of walnuts (Juglans regia L.).\\u003c/em\\u003e Biol Open, 2016. \\u003cstrong\\u003e5\\u003c/strong\\u003e(6): p. 829-36.\\u003c/li\\u003e\\n\\u003cli\\u003eWu, H.C., et al., \\u003cem\\u003eAltered metabolomic-genomic signature: A potential noninvasive biomarker of epilepsy.\\u003c/em\\u003e Epilepsia, 2017. \\u003cstrong\\u003e58\\u003c/strong\\u003e(9): p. 1626-1636.\\u003c/li\\u003e\\n\\u003cli\\u003eDetour, J., et al., \\u003cem\\u003eMetabolomic characterization of human hippocampus from drug-resistant epilepsy with mesial temporal seizure.\\u003c/em\\u003e Epilepsia, 2018. \\u003cstrong\\u003e59\\u003c/strong\\u003e(3): p. 607-616.\\u003c/li\\u003e\\n\\u003cli\\u003eZhang, H., et al., \\u003cem\\u003eUntargeted lipidomic analysis of human hippocampus for temporal lobe epilepsy with hippocampal sclerosis.\\u003c/em\\u003e Epilepsy Res, 2020. \\u003cstrong\\u003e161\\u003c/strong\\u003e: p. 106299.\\u003c/li\\u003e\\n\\u003cli\\u003eFrisardi, V., et al., \\u003cem\\u003eGlycerophospholipids and glycerophospholipid-derived lipid mediators: a complex meshwork in Alzheimer\\u0026apos;s disease pathology.\\u003c/em\\u003e Prog Lipid Res, 2011. \\u003cstrong\\u003e50\\u003c/strong\\u003e(4): p. 313-30.\\u003c/li\\u003e\\n\\u003cli\\u003eXu, S., et al., \\u003cem\\u003eSpatially and temporally probing distinctive glycerophospholipid alterations in Alzheimer\\u0026apos;s disease mouse brain via high-resolution ion mobility-enabled sn-position resolved lipidomics.\\u003c/em\\u003e Nat Commun, 2024. \\u003cstrong\\u003e15\\u003c/strong\\u003e(1): p. 6252.\\u003c/li\\u003e\\n\\u003cli\\u003eXie, J., et al., \\u003cem\\u003eMulti-omics data reveals the important role of glycerophospholipid metabolism in the crosstalk between gut and brain in depression.\\u003c/em\\u003e J Transl Med, 2023. \\u003cstrong\\u003e21\\u003c/strong\\u003e(1): p. 93.\\u003c/li\\u003e\\n\\u003cli\\u003eLi, Z., et al., \\u003cem\\u003eRole of Glycerophospholipid Metabolism in Epilepsy.\\u003c/em\\u003e Curr Neuropharmacol, 2025.\\u003c/li\\u003e\\n\\u003cli\\u003eDahlin, M., C.E. Wheelock, and S. Prast-Nielsen, \\u003cem\\u003eAssociation between seizure reduction during ketogenic diet treatment of epilepsy and changes in circulatory metabolites and gut microbiota composition.\\u003c/em\\u003e EBioMedicine, 2024. \\u003cstrong\\u003e109\\u003c/strong\\u003e: p. 105400.\\u003c/li\\u003e\\n\\u003cli\\u003eChen, Y., et al., \\u003cem\\u003eIntegrated Proteomics and Lipidomics Analysis of Hippocampus to Reveal the Metabolic Landscape of Epilepsy.\\u003c/em\\u003e ACS Omega, 2025. \\u003cstrong\\u003e10\\u003c/strong\\u003e(9): p. 9351-9367.\\u003c/li\\u003e\\n\\u003cli\\u003eKim, H.Y., P.G. Suh, and J.I. Kim, \\u003cem\\u003eThe Role of Phospholipase C in GABAergic Inhibition and Its Relevance to Epilepsy.\\u003c/em\\u003e Int J Mol Sci, 2021. \\u003cstrong\\u003e22\\u003c/strong\\u003e(6).\\u003c/li\\u003e\\n\\u003cli\\u003eQiu, X., et al., \\u003cem\\u003eIntegrative analysis of non-targeted lipidomic data and brain structural imaging identifies phosphatidylethanolamine associated with epileptogenesis.\\u003c/em\\u003e Metabolomics, 2020. \\u003cstrong\\u003e16\\u003c/strong\\u003e(10): p. 110.\\u003c/li\\u003e\\n\\u003cli\\u003evan Echten-Deckert, G., \\u003cem\\u003eThe role of sphingosine 1-phosphate metabolism in brain health and disease.\\u003c/em\\u003e Pharmacol Ther, 2023. \\u003cstrong\\u003e244\\u003c/strong\\u003e: p. 108381.\\u003c/li\\u003e\\n\\u003cli\\u003eBen, X., et al., \\u003cem\\u003eMetabolomics-driven exploration of sphingosine 1-phosphate mechanisms in refractory epilepsy.\\u003c/em\\u003e Neurobiol Dis, 2025. \\u003cstrong\\u003e212\\u003c/strong\\u003e: p. 106953.\\u003c/li\\u003e\\n\\u003cli\\u003eReynolds, E.H., \\u003cem\\u003eAntiepileptic drugs, folate and one carbon metabolism revisited.\\u003c/em\\u003e Epilepsy Behav, 2020. \\u003cstrong\\u003e112\\u003c/strong\\u003e: p. 107336.\\u003c/li\\u003e\\n\\u003cli\\u003eBaxter, M.G., A.A. Miller, and R.A. Webster, \\u003cem\\u003eSome studies on the convulsant action of folic acid.\\u003c/em\\u003e Br J Pharmacol, 1973. \\u003cstrong\\u003e48\\u003c/strong\\u003e(2): p. 350p-351p.\\u003c/li\\u003e\\n\\u003cli\\u003eKim, D.W., et al., \\u003cem\\u003eEffects of new antiepileptic drugs on circulatory markers for vascular risk in patients with newly diagnosed epilepsy.\\u003c/em\\u003e Epilepsia, 2013. \\u003cstrong\\u003e54\\u003c/strong\\u003e(10): p. e146-9.\\u003c/li\\u003e\\n\\u003cli\\u003eNi, G., et al., \\u003cem\\u003eEffects of antiepileptic drug monotherapy on one-carbon metabolism and DNA methylation in patients with epilepsy.\\u003c/em\\u003e PLoS One, 2015. \\u003cstrong\\u003e10\\u003c/strong\\u003e(4): p. e0125656.\\u003c/li\\u003e\\n\\u003cli\\u003eRamaekers, V.T. and E.V. Quadros, \\u003cem\\u003eCerebral Folate Deficiency Syndrome: Early Diagnosis, Intervention and Treatment Strategies.\\u003c/em\\u003e Nutrients, 2022. \\u003cstrong\\u003e14\\u003c/strong\\u003e(15).\\u003c/li\\u003e\\n\\u003cli\\u003eWheless, J.W. and J.M. Rho, \\u003cem\\u003eRole of cholesterol in modulating brain hyperexcitability.\\u003c/em\\u003e Epilepsia, 2025. \\u003cstrong\\u003e66\\u003c/strong\\u003e(1): p. 33-46.\\u003c/li\\u003e\\n\\u003cli\\u003eHawkins, N.A., et al., \\u003cem\\u003eSoticlestat, a novel cholesterol 24-hydroxylase inhibitor, reduces seizures and premature death in Dravet syndrome mice.\\u003c/em\\u003e Epilepsia, 2021. \\u003cstrong\\u003e62\\u003c/strong\\u003e(11): p. 2845-2857.\\u003c/li\\u003e\\n\\u003cli\\u003eNishi, T., et al., \\u003cem\\u003eAnticonvulsive properties of soticlestat, a novel cholesterol 24-hydroxylase inhibitor.\\u003c/em\\u003e Epilepsia, 2022. \\u003cstrong\\u003e63\\u003c/strong\\u003e(6): p. 1580-1590.\\u003c/li\\u003e\\n\\u003cli\\u003eHahn, C.D., et al., \\u003cem\\u003eA phase 2, randomized, double-blind, placebo-controlled study to evaluate the efficacy and safety of soticlestat as adjunctive therapy in pediatric patients with Dravet syndrome or Lennox-Gastaut syndrome (ELEKTRA).\\u003c/em\\u003e Epilepsia, 2022. \\u003cstrong\\u003e63\\u003c/strong\\u003e(10): p. 2671-2683.\\u003c/li\\u003e\\n\\u003cli\\u003eXia, J. and D.S. Wishart, \\u003cem\\u003eWeb-based inference of biological patterns, functions and pathways from metabolomic data using MetaboAnalyst.\\u003c/em\\u003e Nat Protoc, 2011. \\u003cstrong\\u003e6\\u003c/strong\\u003e(6): p. 743-60.\\u003c/li\\u003e\\n\\u003cli\\u003eRobin, X., et al., \\u003cem\\u003epROC: an open-source package for R and S+ to analyze and compare ROC curves.\\u003c/em\\u003e BMC Bioinformatics, 2011. \\u003cstrong\\u003e12\\u003c/strong\\u003e: p. 77.\\u003c/li\\u003e\\n\\u003c/ol\\u003e\"}],\"fulltextSource\":\"\",\"fullText\":\"\",\"funders\":[],\"hasAdminPriorityOnWorkflow\":false,\"hasManuscriptDocX\":true,\"hasOptedInToPreprint\":true,\"hasPassedJournalQc\":\"\",\"hasAnyPriority\":false,\"hideJournal\":false,\"highlight\":\"\",\"institution\":\"\",\"isAcceptedByJournal\":true,\"isAuthorSuppliedPdf\":false,\"isDeskRejected\":\"\",\"isHiddenFromSearch\":false,\"isInQc\":false,\"isInWorkflow\":false,\"isPdf\":false,\"isPdfUpToDate\":true,\"isWithdrawnOrRetracted\":false,\"journal\":{\"display\":true,\"email\":\"info@researchsquare.com\",\"identity\":\"scientific-reports\",\"isNatureJournal\":false,\"hasQc\":true,\"allowDirectSubmit\":false,\"externalIdentity\":\"scirep\",\"sideBox\":\"Learn more about [Scientific Reports](http://www.nature.com/srep/)\",\"snPcode\":\"\",\"submissionUrl\":\"\",\"title\":\"Scientific Reports\",\"twitterHandle\":\"\",\"acdcEnabled\":true,\"dfaEnabled\":true,\"editorialSystem\":\"stoa\",\"reportingPortfolio\":\"Scientific Reports\",\"inReviewEnabled\":true,\"inReviewRevisionsEnabled\":true},\"keywords\":\"epilepsy, drug-refractory epilepsy, serum metabolomics, predictive model\",\"lastPublishedDoi\":\"10.21203/rs.3.rs-7173411/v1\",\"lastPublishedDoiUrl\":\"https://doi.org/10.21203/rs.3.rs-7173411/v1\",\"license\":{\"name\":\"CC BY 4.0\",\"url\":\"https://creativecommons.org/licenses/by/4.0/\"},\"manuscriptAbstract\":\"\\u003cp\\u003eDrug-resistant epilepsy (DRE) lacks a reliable early warning system. This prospective study developed an early predictive model for DRE based on baseline serum metabolite levels. Serum samples were collected from 151 prospectively recruited patients with epilepsy. Patients were categorized into the DRE and non-drug-refractory epilepsy (NDRE) groups after a four-year follow up. After propensity score matching of baseline data, including age and sex, 33 patients with DRE and 89 with NDRE were recruited and split into training and test sets at random in a 2:1 ratio. Nontargeted metabolomics of the training set identified 109 significantly altered metabolites, primarily those involved in dysregulated lipid metabolism. Compared with the NDRE group, 22 metabolites were upregulated and 87 were downregulated in the DRE group. Pathway enrichment analysis highlighted perturbations in choline metabolism in cancer and sphingolipid, glycerophospholipid, one-carbon pool by folate, and cholesterol metabolism. A support vector machine model incorporating 11 metabolites and one clinical characteristic achieved a cross-validated area under the curve (AUC) of 0.810 and an independent test set AUC of 0.794. This study provides a non-invasive, serum-based objective tool to identify the potential population of patients with DRE with good sensitivity and specificity and guide targeted metabolic therapies.\\u003c/p\\u003e\",\"manuscriptTitle\":\"Alterations in serum metabolomics predict drug-resistant epilepsy\",\"msid\":\"\",\"msnumber\":\"\",\"nonDraftVersions\":[{\"code\":1,\"date\":\"2025-08-19 14:24:22\",\"doi\":\"10.21203/rs.3.rs-7173411/v1\",\"editorialEvents\":[{\"type\":\"communityComments\",\"content\":0},{\"type\":\"decision\",\"content\":\"Revision requested\",\"date\":\"2025-09-19T13:34:59+00:00\",\"index\":\"\",\"fulltext\":\"\"},{\"type\":\"editorInvitedReview\",\"content\":\"\",\"date\":\"2025-09-13T09:04:16+00:00\",\"index\":\"hide\",\"fulltext\":\"\"},{\"type\":\"reviewerAgreed\",\"content\":\"22283193422480677347674996217694596986\",\"date\":\"2025-09-03T11:16:14+00:00\",\"index\":\"hide\",\"fulltext\":\"\"},{\"type\":\"editorInvitedReview\",\"content\":\"\",\"date\":\"2025-08-27T16:15:02+00:00\",\"index\":\"hide\",\"fulltext\":\"\"},{\"type\":\"reviewerAgreed\",\"content\":\"140893116216273641308591105676288688180\",\"date\":\"2025-08-25T13:56:01+00:00\",\"index\":\"hide\",\"fulltext\":\"\"},{\"type\":\"reviewersInvited\",\"content\":\"\",\"date\":\"2025-08-11T22:25:47+00:00\",\"index\":\"\",\"fulltext\":\"\"},{\"type\":\"editorInvited\",\"content\":\"\",\"date\":\"2025-07-23T15:13:24+00:00\",\"index\":\"\",\"fulltext\":\"\"},{\"type\":\"editorAssigned\",\"content\":\"\",\"date\":\"2025-07-23T07:01:28+00:00\",\"index\":\"\",\"fulltext\":\"\"},{\"type\":\"checksComplete\",\"content\":\"\",\"date\":\"2025-07-23T01:04:12+00:00\",\"index\":\"\",\"fulltext\":\"\"},{\"type\":\"submitted\",\"content\":\"Scientific Reports\",\"date\":\"2025-07-21T05:07:37+00:00\",\"index\":\"\",\"fulltext\":\"\"}],\"status\":\"published\",\"journal\":{\"display\":true,\"email\":\"info@researchsquare.com\",\"identity\":\"scientific-reports\",\"isNatureJournal\":false,\"hasQc\":true,\"allowDirectSubmit\":false,\"externalIdentity\":\"scirep\",\"sideBox\":\"Learn more about [Scientific Reports](http://www.nature.com/srep/)\",\"snPcode\":\"\",\"submissionUrl\":\"\",\"title\":\"Scientific Reports\",\"twitterHandle\":\"\",\"acdcEnabled\":true,\"dfaEnabled\":true,\"editorialSystem\":\"stoa\",\"reportingPortfolio\":\"Scientific Reports\",\"inReviewEnabled\":true,\"inReviewRevisionsEnabled\":true}}],\"origin\":\"\",\"ownerIdentity\":\"80959576-a0e4-4c16-ab14-1a6e9c352e33\",\"owner\":[],\"postedDate\":\"August 19th, 2025\",\"published\":true,\"recentEditorialEvents\":[],\"rejectedJournal\":[],\"revision\":\"\",\"amendment\":\"\",\"status\":\"published-in-journal\",\"subjectAreas\":[{\"id\":53153250,\"name\":\"Health sciences/Biomarkers\"},{\"id\":53153251,\"name\":\"Biological sciences/Cancer\"},{\"id\":53153252,\"name\":\"Health sciences/Diseases\"},{\"id\":53153253,\"name\":\"Health sciences/Medical research\"}],\"tags\":[],\"updatedAt\":\"2025-12-08T16:09:53+00:00\",\"versionOfRecord\":{\"articleIdentity\":\"rs-7173411\",\"link\":\"https://doi.org/10.1038/s41598-025-30741-8\",\"journal\":{\"identity\":\"scientific-reports\",\"isVorOnly\":false,\"title\":\"Scientific Reports\"},\"publishedOn\":\"2025-12-03 15:58:04\",\"publishedOnDateReadable\":\"December 3rd, 2025\"},\"versionCreatedAt\":\"2025-08-19 14:24:22\",\"video\":\"\",\"vorDoi\":\"10.1038/s41598-025-30741-8\",\"vorDoiUrl\":\"https://doi.org/10.1038/s41598-025-30741-8\",\"workflowStages\":[]},\"version\":\"v1\",\"identity\":\"rs-7173411\",\"journalConfig\":\"researchsquare\"},\"__N_SSP\":true},\"page\":\"/article/[identity]/[[...version]]\",\"query\":{\"redirect\":\"/article/rs-7173411\",\"identity\":\"rs-7173411\",\"version\":[\"v1\"]},\"buildId\":\"8U1c8b4HqxoKbykW_rLl7\",\"isFallback\":false,\"isExperimentalCompile\":false,\"dynamicIds\":[84888],\"gssp\":true,\"scriptLoader\":[]}","source_license":"CC-BY-4.0","license_restricted":false}