Gut microbiota shifts and short‐chain fatty acids alterations in pediatric epilepsy patients on a Mediterranean ketogenic diet

Abstract

Objective

The olive oil–based Mediterranean ketogenic diet (MedKD) may support patients with drug‐resistant epilepsy (DRE) or neurometabolic disorders by integrating ketogenic therapy with the cardiometabolic and neuroprotective advantages of the Mediterranean diet. This study assessed alterations in gut microbiota composition and short‐chain fatty acids (SCFAs) in Greek pediatric epilepsy patients under a 3‐month MedKD.

Methods

Patients eligible for ketogenic diet therapy and one parent per patient as controls were enrolled. An olive oil–based KD aligned with Mediterranean principles was initiated during a 5–7‐day hospital stay. Anthropometric, nutritional, biochemical and clinical data were recorded at baseline and after 3 months. Fecal samples were collected from patients at both time points, and parents provided a single baseline sample. Microbiota composition was analyzed using 16S rRNA gene next‐generation sequencing, and SCFAs were quantified via gas chromatography. Written informed consent was obtained from all parents. The study was registered at ClinicalTrials.gov (NCT05898438).

Results

The study enrolled 18 pediatric/adolescent patients (aged 2.5–15.5 years) and 17 parents at baseline (T1). After 3 months (T2), 13 patient–parent pairs completed follow‐up. MedKD adherence was high, with mean ketone levels of 2.6 mmol/L and a clinical response in 85% of patients. Anthropometric measurements improved, and lipid changes were modest. No significant α‐diversity differences were detected between patients and parents at T1 or between T1 and T2 in patients, whereas β‐diversity differed significantly between parents and patients at baseline. Epilepsy‐associated genera were more abundant in patients. At T2, patients exhibited shifts including increased members of Eggerthellaceae and decreased Lachnospira and Bifidobacterium. Major SCFAs remained stable, with increases limited to minor protein‐derived SCFAs.

Significance

The MedKD was associated with high adherence, efficacy, and anthropometric improvements, along with modest effects on the lipid profile. Microbial and SCFA changes reflected both the epilepsy‐related milieu and potential influences of the diet's high olive oil content.

Key points.

• The Mediterranean ketogenic diet may offer high adherence and efficacy to pediatric/adolescent patients with epilepsy.

• Significantly different age‐ and disease‐related β‐diversity was observed between patients with epilepsy and their healthy parents.

• The microbial shifts after the MedKD may be related to the high efficacy observed.

• The unchanged key SCFA levels may be related to the phenolic components of olive oil provided with the diet.

1. INTRODUCTION

The ketogenic diet (KD) is a well‐established therapeutic option for children and adults with drug‐resistant epilepsy (DRE) or certain metabolic disorders. ¹ This high‐fat, low‐carbohydrate diet reduces seizures and influences cognition and behavior through multiple mechanisms, including ketone body production, altered glucose metabolism, improved mitochondrial function, anti‐inflammatory and antioxidative pathways, epigenetic regulation, and gut microbiome. ²

Through effects on the “gut–brain axis,” KD‐induced microbial changes may contribute to improved brain function and overall health in DRE patients. ³

To improve adherence and tolerability, several KD variants have been developed, which may differentially affect gut microbiota given the strong influence of diet on microbial composition. ⁴ The Mediterranean ketogenic diet (MedKD) combines KD principles with Mediterranean diet characteristics, emphasizing olive oil, fish, nuts, and fruits and vegetables. ⁵ Olive oil, in particular, has been shown to beneficially modulate gut microbiota. ⁶ , ⁷ , ⁸ , ⁹ , ¹⁰

In regions such as Greece, where olive oil is the primary dietary fat, the MedKD may uniquely shape the gut microbiome. ¹¹ To our knowledge, this is the first study examining gut microbiota alterations and fecal short‐chain fatty acids (SCFAs) in Greek pediatric epileptic patients following a MedKD.

2. MATERIALS AND METHODS

2.1. Participants

This observational study included children/adolescents with epilepsy treated with MedKD at the Pediatric Neurology Department, third Pediatric Ward, General University Hospital Attikon (Athens, Greece) from January 2020 to June 2024.

Inclusion criteria were (a) age 2–18 years, (b) diagnosis of DRE per ILAE criteria or epilepsy syndrome requiring KD, (c) no antibiotic use within 2 months before fecal sampling, (d) no probiotic/prebiotic use within 2 weeks before sampling, (e) absence of systemic disease, and (f) no special diet before KD initiation.

2.2. Ethics

The study was approved by the Ethics Committee of Attikon University Hospital (10th Meeting/December 18, 2019) and received a Clinical Trials Number NCT05898438. Parents provided a written informed consent before the intervention.

2.3. Study design—MedKD implementation

Patients were hospitalized for diet initiation. At baseline (T1), blood, urine, and fecal samples were collected, while one parent per patient provided a fecal sample as control. Parents also completed a 3‐day food diary to document baseline dietary intake.

An experienced dietitian developed individualized MedKD plans based on anthropometric and nutritional assessments. The growth status of patients was assessed using Seca scales and stadiometer or a tape (for non‐ambulatory patients) for weight and height measurements. Body mass index (BMI) was calculated as weight (kg) divided by height (m²) and patients were classified as normal, underweight, or overweight using the extended international (IOTF‐International Obesity Task Force) body mass index cutoffs for thinness, overweight, and obesity. ¹²

The diet, characterized by a ketogenic ratio of 2:1–4:1 and >80% of total energy derived from fat (with ideally >50% from olive oil), 10%–15% protein, and 5%–10% carbohydrates, was intended to induce therapeutic ketosis, improve baseline nutritional status, and provide additional health benefits associated with olive oil, a key component commonly recommended in our standard dietary care. Underweight patients were recommended an energy‐enriched regimen to promote growth and weight gain. Patients followed the MedKD for 3 months, emphasizing olive oil, nuts, vegetables, and fish within a Mediterranean‐style approach, without a fasting protocol. At 3 months (T2), anthropometric measurements, blood, urine, and fecal samples were also collected.

Adherence was monitored through regular β‐hydroxybutyrate (β‐HB) and glucose measurements during hospitalization and at home. All patients received age‐appropriate sugar‐free vitamin and mineral supplements.

MedKD efficacy was defined as >50% seizure reduction, EEG improvement (in ESES), or cognitive/motor improvement in nonepileptic conditions (in GM1 gangliosidosis type 2). Parents were instructed to record seizure frequency and observations related to changes in behavior, cognition, and motor function in relevant diaries.

2.4. Blood, urine, and fecal sampling

During hospitalization, patients provided blood and urine samples for routine screening at both T1 and T2. Blood biochemical parameters included blood cells, glucose, urea, creatinine, uric acid, lipids, liver enzymes, electrolytes, minerals, and vitamins. Urine analysis was performed to determine 24‐h calcium and creatinine levels.

2.5. Gut microbiota analysis

Fecal samples were collected from patients at two time points and from parents at T1. Preweighed containers were used for full evacuations. Samples were stored at −80°C or temporarily at −20°C for up to 1 day before transfer to −80°C.

2.5.1. DNA extraction

Fecal samples were allowed to thaw for approximately 2.5 h before being homogenized and weighed. Genomic DNA was extracted from fecal samples by the RBB + C (repeated bead beading plus column) method, after Salonen et al. modifications. ¹³ , ¹⁴ Extracted DNA was checked (Implen photometer) and kept at −80°C until further analysis.

2.5.2. 16S rRNA sequencing

The 16S rRNA gene amplification was performed using a two‐step PCR protocol as outlined in the Illumina application note with primers 341F (5′‐CCTACGGGNGGCWGCAG‐3′) and 785R (5′‐GACTACHVGGGTATCTAATCC‐3′) ¹⁵ targeting the hypervariable V3–V4 region of the 16S rRNA gene. To ensure the reliability of the amplification and sequencing processes, a positive control (ZymoBIOMICS Microbial Community DNA Standard; Zymo Research, Irvine, CA, USA) was included. Additionally, negative controls (PCR‐grade water) were processed alongside the experimental samples to monitor potential contamination (e.g., taxa present in the DNA elution buffer). Sequencing analysis was performed on the Illumina NextSeq 2000 System (paired‐end, 2 × 300 bp) (Illumina Inc., CA, USA) according to the manufacturer's protocol.

2.5.3. Bioinformatics analysis

The raw FASTQ files were processed through the IMNGS platform, ¹⁶ implementing the UPARSE ¹⁷ algorithm from the USEARCH11 package, using the default parameters. The produced denoised sequences were clustered at 97% sequence similarity using the UPARSE algorithm ¹⁷ and then were aligned and classified using SILVA (release 138). ¹⁸ The downstream analysis was performed in R programming language (version 4.4.0) using the Rhea pipeline (version 1.1.6). ¹⁹ For the α‐ and β‐diversity analysis, the raw counts were normalized so that the sum of their counts was equal across all the samples. The α‐diversity was measured in terms of richness and effective richness, ¹⁹ whereas the β‐diversity was calculated using Generalized Unifrac, ²⁰ setting the parameter α = .5 and visualizing the results through the form of MDS plots. For the taxa composition comparisons, the raw counts were transformed into their relative abundances and then compared against each other. The distances‐based part of the analysis and the comparisons between the reference and test groups were performed using DivCom. ²¹

2.6. SCFA analysis

Fecal SCFA analysis was performed as previously described by capillary gas chromatography (GC) (Agilent 6890 GC System, Agilent Technologies, Santa Clara, CA, USA), after 1:3 dilution of fecal samples with .9% saline. ²² , ²³ , ²⁴ Total volatile fatty acids (VFAs) and individual SCFA concentrations were expressed in μmol/g of fecal sample. Molar ratios (% of total VFAs) were calculated for acetate, propionate, butyrate, branched‐chain SCFAs (BSCFAs) including isobutyrate, isovalerate, and isocaproic acid, as well as other SCFAs such as valerate, caproic acid, and heptanoic acid. Fecal pH and stool moisture were also determined. ²² , ²⁵

2.7. Statistical analysis

Continuous variables were reported as median and Q1–Q3 percentiles, and categorical variables as frequencies (n, %). Differences in continuous variables between parents and patients at baseline (T1) were assessed using the Mann–Whitney U test, applying a .5% cutoff for microbial relative abundances. Within‐patient comparisons between T1 and T2 were evaluated using the Wilcoxon‐signed rank test without relative abundance cutoff filtering. Categorical variables were analyzed using the chi‐squared test.

β‐diversity differences were assessed by PERMANOVA, and group homogeneity by PERMDISP, using the vegan package in R. ²⁶ , ²⁷ Statistical analyses were performed in Stata 15.1. ²⁸ p‐values were adjusted using the Benjamini–Hochberg method, with adjusted p < .05 considered statistically significant.

3. RESULTS

3.1. Cohort characteristics

A total of 18 children/adolescents and 17 parents provided fecal samples before MedKD initiation (T1) (Figure 1). Patients 2–14 (n = 13) (Table 1) provided suitable fecal samples at T2 after the 3‐month intervention—this final cohort included 5 males (61.5%) (mean age 7.6 years) and 8 females (38.5%) (mean age 6.75 years) with multiple epilepsy etiology and varying numbers of AEDs. Patients' characteristics are shown in Tables 1 and 2.

FIGURE 1.

Flow diagram of the study.

TABLE 1.

Cohort characteristics, diagnosis, and number of AEDs.

N	Sex	Diagnosis	Age at diagnosis (years)	Age at T1 (years)	Number of AEDs at T1
1	F	Encephalopathy ESES	2.5	16.5	3
2	F	Autoimmune encephalitis (anti‐GAD)	3	15.5	2
3	M	Encephalopathy ESES	6.5	8.5	3
4	F	Epileptic encephalopathy (SCN1B)	1.5	9.5	3
5	M	Epileptic encephalopathy (CDKL‐5)	.2	2.5	2
6	F	Drug resistant epilepsy beginning with Infantile spasms	.9	3	2
7	F	Myoclonic absence epilepsy	5	9.5	2
8	F	Drug resistant epilepsy beginning with Infantile spasms	.5	3.5	3
9	F	Lissencephaly (grade 3)	.4	8	4
10	M	Doose syndrome	2.5	5.5	3
11	F	Epileptic encephalopathy (CDKL‐5)	1.5	2.5	3
12	M	GM1 Gangliosidosis type 2	12	12	1
13	F	Epileptic encephalopathy (CDKL‐5)	1	2.5	3
14	M	GM1 Gangliosidosis type 2	8	9.5	1
15	F	Drug resistant epilepsy	8	16.5	2
16	M	GLUT 1 Deficiency Syndrome	3	3.5	1
17	M	Epileptic encephalopathy	4.5	7	2
18	M	Drug resistant epilepsy	.9	2	2

Abbreviations: AED, antiepileptic drugs; CDKL‐5, cyclin‐dependent kinase‐like 5 (deficiency disorder); ESES, electrical status epilepticus during sleep; F, female; GAD, glutamic acid decarboxylase; GLUT, glucose transporter; M, male; SCN1B, sodium channel gene variant.

TABLE 2.

Patients' anthropometric characteristics and MedKD analysis ^a .

	N	T1	T2	p‐Value
Age	7.07 (SD: 4.18) (n = 13)
Sex
Female	8/13 (61.5%)
Male	5/13 (38.5%)
BMI (kg/m2)		14.80 (13.70, 16.03)	15.92 (14.53, 17.71)	.011
Weight status (%)
Healthy weight		53.8	69.2	.64
Underweight		38.5	23.1
Overweight		7.7	7.7
Energy (kcal/day)		1378.01 (1139.42, 1599.27)	1556.80 (1211.69, 1866.02)	.196
Carbohydrates (g/day)		92.43 (62.23, 131.04)	24.21 (18.97, 34.39)	.002
Total fiber (g/day)		10.22 (5.94, 14.81)	4.62 (2.35, 8.21)	.016
Proteins (g/day)		64.26 (47.05, 86.17)	32.92 (30.30, 47.86)	.003
Fat (g/day)		78.86 (61.58, 103.75)	146.62 (112.24, 172.25)	.002
%Fat as olive oil		29.0 (24.0, 38.5)	58.7 (50.7, 67.5)	.001

Abbreviation: BMI, body mass index.

^a Values expressed as median and interquartile percentiles (Q1, Q3).

Significant increase in BMI was observed among the 13 patients who completed the study (p < .05), with the median BMI increasing from 14.8 kg/m² at T1 to 15.92 kg/m² at T2 (Table 2). Among the patients, 53.8% (7/13) had normal weight at baseline, with a trend to increase after MedKD. Prior the diet initiation, five patients (38.5%) were underweight, resulting in significant BMI increase after MedKD, with 2 of them resulting from moderate to slightly underweight and 2 others to normal weight classification. One patient (7.7%) exhibited an increase in BMI of 2.10 units following the dietary intervention, further reinforcing the initial classification of marginal overweight. Overall, at T2, the proportion of patients with normal weight increased from 53.8% to 69.2%, whereas the proportion of underweight patients decreased from 38.5% to 23.1%.

3.2. MedKD characteristics

As shown in Table 2, the median total energy intake was higher at T2 (1557 kcal) compared to T1 (1378 kcal). Carbohydrate intake was lower at T2 (median 24 g) than at T1 (92 g), while protein intake also decreased from 64 g at T1 to 33 g at T2. Olive oil contributed nearly 60% of the total fat intake, highlighting its central role in the dietary pattern.

The ketogenic ratio (fat to combined protein and carbohydrate) varied between 1.7:1 and 4:1, with a mean of approximately 2.4:1. The lowest ketogenic ratio used was 1.7:1 for patient 3, due to the easily achievable high ketosis. The highest ratios, 3.7:1 and 4:1, were prescribed for patients 3 and 5, respectively. The remaining patients adhered to a MedKD regimen with ketogenic ratios ranging from 2:1 to 2.6:1. All patients received oral nutrition. None were administered enteral feeding via nasogastric tube or gastrostomy and no commercially available enteral formulas were utilized.

3.3. KD adherence

The β‐HB blood levels were used as a measure of adherence. Of the 13 patients, one had β‐ΗΒ levels at .9 mmol/L, five had levels between 2.0 and 2.5 mmol/L, and seven had levels between 3.0 and 3.5 mmol/L. The mean β‐HB value was 2.6 mmol/L (SD .7 mmol/L). Overall, patients' adherence was high, with only minor issues observed during the initial days of MedKD treatment.

3.4. MedKD efficacy

Two patients (4 and 5) were classified as nonresponders, showing effects only in the first month of MedKD treatment. The remaining 11 responders had improvements in behavioral and motor functions, as documented in parental diaries and clinical examination reports.

Patients 12 and 14 (GM1 gangliosidosis) improved in motor function and alertness, while patient 3 (ESES) experienced an improved Spike and Wave Index. Patient 7 became seizure‐free; patients 2 and 13 had >90% seizure reduction; and patients 6, 8, 9, 10, and 11 showed a 50%–90% seizure reduction. Seizure reduction was also assessed based on parental diaries.

3.5. Blood sample analysis

Low‐density cholesterol levels (LDL‐C), uric acid and sodium levels significantly increased, whereas serum glutamic oxaloacetic transaminase (SGOT), white blood cells (WBC), urea, and amylase significantly decreased (p < .05) after the 3‐month MedKD implementation. Total cholesterol marginally increased at T2 (p = .050). No significant changes were detected in fasting blood glucose, triglycerides, high‐density cholesterol levels (HDL‐C), and serum glutamic pyruvic transaminase (SGPT) concentrations (p > .05) (Table S1).

3.6. Fecal samples analyses

3.6.1. Parents and patients at T1

Alpha (α) and beta (β) diversity

Microbial α‐diversity at baseline was assessed using Shannon index, Simpson index, and normalized richness metrics (Table S2). None of these showed significant differences between parents (P) and patients (T1) at baseline (Figure 2A).

FIGURE 2.

α‐diversity (A) and β‐diversity (B) indices of parents (P) (n = 17) and patients (T1) (n = 18) at baseline.

To assess differences in overall gut–microbiota composition between groups, beta‐diversity analysis was performed using both PERMANOVA and PERMDISP tests (Figure 2B). Comparison between parents and patients revealed a significant difference in gut–microbial community structure at baseline, with a distinct clustering of microbial profiles between the two groups (PERMANOVA analysis p = .033, Figure 2B). Furthermore, PERMDISP analysis showed no significant difference in the dispersion of microbial communities between groups (p = .408), suggesting that the observed differences in β‐diversity are driven by differences in the overall microbial community composition (group centroids) rather than variation of the communities within each group (Figure 2B).

Taxonomic analysis

Analysis of microbial community composition through taxonomic binning at phyla level along with their cumulative relative abundances is presented in Figure 3. The dominant phyla were Bacillota (or Firmicutes) and Bacteroidota (Bacteroidetes) in both parents and patients at T1 (Bacillota: parents 62%, patients T1: 58.5%—p = .129, Bacteroidota: parents 19%, patients T1: 19%—p = .741).

FIGURE 3.

Taxonomic binning at Phyla level of parents (P) (n = 17) and patients (T1) (n = 18) at baseline.

Significantly higher relative abundances of Bacilli at the class level, Lactobacillales at the order level, and Streptococcaceae at the family level were observed in patients compared with parents at T1 (Table 3). A nonsignificant trend toward higher levels in patients was observed for the order Clostridiales, the family Clostridiaceae, the genus Clostridium, and the genus Streptococcus (Table S3).

TABLE 3.

Significant differences in gut community between 17 parents (P) and 18 patients (T1) at baseline ^a .

Taxa	Patients	Parents	p‐Value
Taxa higher in patients
c_Bacilli	10.99 (5.30–15.74)	5.01 (3.20–8.73)	.045
o_Lactobacillales	7.03 (2.78–13.77)	1.41 (0.80–2.88)	.002
f_Streptocaccaceae	5.91 (2.11–8.60)	1.31 (0.58–2.07)	.020
Taxa lower in patients
c_Clostridia	45.01 (32.63–52.36)	53.41 (49.94–59.57)	.004
o_Oscillospirales	13.96 (8.51–19.78)	22.56 (16.17–27.57)	.003
f‐Ruminococcaceae	10.49 (3.67–13.91)	15.96 (12.85–19.82)	.001
g‐Ruminococcus	1.23 (0.92–2.46)	4.76 (1.97–8.93)	.005
g_Unknown Lachnospiraceae	3.91 (1.88–4.25)	5.48 (3.57–7.44)	.005

^a Values expressed as median relative abundance (%) and interquartile percentiles (Q1–Q3).

In contrast, patients exhibited significantly lower relative abundances than parents at T1 for the class Clostridia, the order Oscillospirales, the family Ruminococcaceae, as well as the genera Ruminococcus and unclassified Lachnospiraceae (Table 3). Additional taxa showed a nonsignificant trend toward lower abundance in patients at T1, such as the family Coriobacteriaceae (Table S3).

Using the cutoff value of .5% in microbial relative abundances, higher detection frequency (Fisher's exact test applied) was observed in patients compared to parents (11 vs. 5) for the phylum Pseudomonadota (p = .0018), the order Enterobacterales (11 vs. 4) (p = .0409), the family Enterobacteriaceae (11 vs. 2) (p = .0045), and the genus Esherichia‐Shigella (10 vs. 2) (p = .0116), whereas lower frequency was observed in patients for the genus unknown Ruminococcaceae (11 vs. 16) (p = .0408).

SCFAs analysis

Concentrations (μmol/g of feces) and molar ratios (% of TVFAs) of fecal SCFAs for parents and patients at T1 are presented in Table 4.

TABLE 4.

Fecal SCFAs concentrations (μmol/g of feces) and molar ratios (%TVFAs) of parents (n = 17) and patients (n = 18) at T1 ^a .

Fecal SCFAs	Patients	Parents	p‐Value	Patients	Parents	p‐Value
	Concentration, μmol/g			Molar ratio, %
Acetate	29.47 (19.56–46.18)	28.65 (21.21–43.20)	.869	52.51 (50.13–55.36)	50.29 (46.01–55.43)	.210
Propionate	9.96 (5.14–17.91)	11.82 (6.41–19.75)	.409	17.88 (11.91–21.98)	20.18 (17.25–23.30)	.210
Isobutyrate	1.21 (0.46–2.62)	1.14 (0.73–2.33)	.974	2.71 (1.09–3.88)	3.03 (1.38–3.50)	.717
Butyrate	11.13 (4.22–16.80)	11.49 (6.21–16.14)	.792	16.16 (11.92–22.82)	18.37 (13.87–21.68)	.488
Isovalerate	1.94 (0.99–2.62)	1.82 (1.27–4.23)	.552	3.45 (1.73–5.92)	5.36 (2.06–6.83)	.306
Valerate	1.26 (0.76–2.18)	1.54 (1.08–2.81)	.222	2.50 (1.06–3.73)	3.43 (2.02–4.43)	.222
Isocaproic	0.00 (0.00–0.06)	0.00 (0.00–0.00)	.059	0.00 (0.00–0.06)	0.00 (0.00–0.00)	.053
Caproic	0.07 (0.00–0.11)	0.13 (0.03–0.73)	.081	0.09 (0.00–0.22)	0.32 (0.05–1.14)	.035
Heptanoic	0.00 (0.00–0.00)	0.00 (0.00–0.10)	.043	0.00 (0.00–0.00)	0.00 (0.00–0.18)	.043
BSCFAs	3.31 (1.45–5.66)	2.82 (2.26–6.62)	.869	6.90 (3.21–10.59)	8.06 (3.60–10.37)	.895
Other SCFAs	1.44 (0.76–2.18)	1.93 (1.17–3.30)	.166	2.70 (1.12–4.43)	4.23 (2.26–5.33)	.137
TVFAs	60.60 (32.55–87.97)	55.68 (36.17–89.95)	.974

Abbreviations: BSCFAs, branched short‐chain fatty acids; SCFAs, short‐chain fatty acids; TVFAs, total volatile fatty acids.

^a Values expressed as median and interquartile percentiles (Q1–Q3).

Concentrations and molar ratios of acetate, propionate, butyrate, and valerate had no significant difference between parents and patients at T1 (all p > .05). Acetate was the most abundant SCFA in both groups. Significant lower proportions of heptanoic and caproic acids and a trend for higher isocaproic acid molar ratio were observed in patients compared to parents at T1. Total volatile fatty acids concentrations were similar for both patients and parents at baseline (Table 4).

3.6.2. Patients at T1 and T2

Alpha (a) and beta (β) diversity

The α‐diversity in fecal samples from 13 patients remained stable between T1 and T2 during the MedKD intervention, with no significant differences observed, although a trend toward increased richness at T2 was noted (p < .1) (Figure 4A) (Table S4). Furthermore, β‐diversity analysis showed no significant change in microbial community composition or dispersion after the MedKD intervention (Figure 4B).

FIGURE 4.

Alpha diversity (A) and beta diversity (B) of patients at T1 and T2 (n = 13).

Taxonomic analysis

The taxonomic binning revealed no significant microbial shifts at phyla level between timepoints T1 and T2 (Figure 5). Nevertheless, trends for increased relative abundance of Bacillota (59.87% vs. 64.64%, p = .152) and decreased relative abundance of Actinomycetota (13.32% vs. 8.86%, p = .087) and Bacteroidota (21.33% vs. 12.12%, p = .1173) were indicated after the MedKD implementation (Figure 5).

FIGURE 5.

Taxonomic binning at Phyla level in patients at T1 and T2 (n = 13).

Several relative abundances of microbial genera demonstrated significant increases postdiet, including Romboutsia and unknown members of Eggerthellaceae. Other notable increases included Ruthenibacterium, Coprobacillus, Extibacter, Gordonibacter, and Eisenbergiella (Table 5). Conversely, members of the Lachnospiraceae family, including the genera Lachnospiraceae ND3007 group and Lachnospira, exhibited significant reductions in relative abundance under MedKD. Similarly, the abundance of Bifidobacterium and related taxa within the Bifidobacteriaceae family and Bifidobacteriales order decreased significantly, along with Veillonella and Haemophilus. Additionally, Haemophilus detection frequency was also decreased (from 10 at T1 to 4 patients at T2), under no cutoff filtering in relative abundances (Table 5).

TABLE 5.

Significant differences in gut community at genus level in patients between T1 and T2 ^a .

Genera	Patients T1	Patients T2	p‐Value
Increased genera at T2
Clostridium innocuum group	.0179 (.0078–.0972)	.0844 (.0282–.1800)	.033
Dielma	.0005 (.0000–.0049)	.0047 (.0000–.0174)	.033
Thomasclavelia	.1032 (.0083–1.6141)	.4843 (.0598–2.5690)	.031
Romboutsia	.0168 (.0003–.1322)	.2219 (.0454–.9777)	.008
Unknown Eggerthellaceae	.0015 (.0000–.0750)	.0118 (.0000–.1424)	.008
Extibacter	.0000 (.0000–.0061)	.0322 (.0070–.1555)	.037
Coprobacillus	.0016 (.0002–.0126)	.0119 (.0000–.0613)	.018
Eisenbergiella	.0037 (.0000–.0211)	.0416 (.0028–.1886)	.018
Ruthenibacterium	.0636 (.0099–.1233)	.2140 (.0400–.3691)	.017
Gordonibacter	.0060 (.0015–.0201)	.0072 (.0007–.0406)	.023
Epulopiscium	.0004 (.0000–.0029)	.0020 (.0000–.0277)	.041
Schaalia	.0100 (.0013–.0281)	.0185 (.0055–.1247)	.045
Actinomyces	.0037 (.0017–.0075)	.0059 (.0021–.0134)	.048
Decreased genera at T2
Lachnospiraceae ND3007 group	.1223 (.0335–.2620)	.0056 (.0000–.0429)	.006
Bifidobacterium	8.7389 (2.9326–18.330)	1.7610 (.3036–10.3611)	.028
Lachnospira	.2198 (.0836–.3503)	.0478 (.0142–.1569)	.046
Veillonella	.0732 (.0114–.8309)	.0039 (.0008–.2768)	.041
Haemophilus	.0044 (.0011–.0229)	.0000 (.0000–.0080)	.035

^a Values expressed as median relative abundance (%) and interquartile percentiles (Q1–Q3).

Other taxa showed a trend to increased levels at T2 (p < .1), such as the families Peptostreptococcaceae and Christensenellaceae and the genera Flavonifractor and Enterococcus, whereas other taxa exhibited a reduction trend at T2 such as the family Pasteurellaceae (p < .1) and the genus Lachnospiraceae NK4A136 group (Table S5).

SCFAs analysis

No significant differences in the concentrations and molar ratios of SCFAs were observed between time points T1 and T2 for the 13 patients (all p > .05) (Table 6), with a trend for increased molar ratio of isocaproic acid at T2 (p = .063) and a trend for lower concentrations of propionate (p = .101) and TVFAs at T2 compared to T1 (p = .055).

TABLE 6.

Fecal SCFAs concentrations (μmol/g) and molar ratios (% of TVFAs) of the final cohort at T1 and T2 ^a .

SCFA	Patients T1	Patients T2	p‐Value	Patients T1	Patients T2	p‐Value
	Concentration, μmol/g			Molar ratio, %
Acetate	29.69 (27.48–46.86)	29.96 (24.20–39.05)	.311	52.90 (50.88–61.67)	59.12 (54.89–64.68)	.279
Propionate	9.41 (4.83–19.76)	7.57 (5.29–0.85)	.101	13.43 (11.28–20.30)	15.38 (12.89–17.31)	.600
Isobutyrate	1.16 (.43–2.19)	1.44 (.49–2.11)	.972	1.79 (.82–3.88)	2.45 (1.34–4.23)	.753
Butyrate	11.66 (5.74–17.58)	8.49 (4.91–13.01)	.116	16.36 (10.54–26.28)	14.95 (12.10–18.83)	.382
Isovalerate	1.92 (.92–2.74)	2.81 (1.04–3.58)	.221	2.50 (1.09–5.65)	4.15 (2.89–7.21)	.116
Valerate	1.21 (.59–1.95)	1.51 (.28–1.98)	.972	2.07 (.60–3.14)	2.45 (.76–3.49)	.279
Isocaproic	.00 (.00–.06)	.00 (.00–.07)	.091	.00 (.00–.05)	.00 (.00–.14)	.063
Caproic	.07 (.00–.11)	.00 (.00–.27)	.859	.08 (.00–.14)	.00 (.00–.47)	.594
Heptanoic	.00 (.00–.00)	.00 (.00–.03)	.109	.00 (.00–.03)	.00 (.00–.08)	.109
BSCFAs	3.14 (1.34–6.00)	4.20 (1.54–5.73)	.807	4.88 (2.12–10.61)	6.74 (4.25–11.44)	.422
Other SCFAs	1.30 (.65–1.95)	1.66 (.28–2.16)	.917	2.15 (.71–3.23)	2.72 (1.31–3.71)	.345
TVFAs	64.94 (46.36–100.90)	52.30 (41.15–71.45)	.055

^a Values expressed as median and interquartile percentiles (Q1–Q3).

4. DISCUSSION

The KD has been extensively utilized in pediatric patients with DRE, with emerging research examining its influence on the gut microbiota and its potential implications for precision medicine. ²⁹ , ³⁰ , ³¹ , ³² , ³³ In this study, 85% of pediatric epilepsy patients on an olive oil–rich MedKD showed a positive therapeutic response, reflecting high efficacy and adherence. Sustained ketone levels over 3 months and a BMI increase further indicated compliance. Macronutrient analysis confirmed consistent olive oil intake, accounting for approximately 60% of total fat, likely aided by familiarity with the Mediterranean diet, as reported in previous MedKD studies. ³⁴ , ³⁵ , ³⁶ , ³⁷

The MedKD, although rich in olive oil, resulted in an elevation of LDL cholesterol at T2, with a marginally significant increase in total cholesterol, as also reported in other studies exploring the KD effect on lipid profiles. ³⁸ , ³⁹ , ⁴⁰ However, the observed increases in LDL‐C and total cholesterol were modest, possibly due to the high content of MUFAs and phenolic compounds from the olive oil‐rich diet. ⁴¹

In the present study, α‐diversity did not differ significantly between parents and patients at baseline, consistent with the findings of Wang et al., who compared patients with mitochondrial epilepsy to healthy controls (HC). ⁴² In contrast, other studies have reported lower α‐diversity in patients with DRE compared with HC, whereas Gong et al. observed higher α‐diversity in DRE patients. ²⁹ , ³⁰ , ³³

Analysis of β‐diversity revealed distinct, potentially age‐ and disease‐related microbial profiles between patients and parents, as also reported by Lindefeldt et al. in their similar design study. ³⁰

At the phylum level, no significant differences in relative abundances were observed between patients and parents at T1. On the contrary, Lindefeldt et al. reported taxonomic shifts at phylum level in children with epilepsy compared to parental controls, including a reduction in Actinobacteria (Actinomycetota) and Bacillota (Firmicutes), and an increase in Proteobacteria (Pseudomonadota) and Bacteroidetes within the patient. ³⁴ Other researchers also demonstrate differences regarding the abundance of Bacillota, Actinobacteria, Verrucomicrobia, Pseudomonadota, and Bacteroidota between epileptic patients (pediatric or adults) and healthy siblings. ²⁹ , ³¹ , ³³ , ⁴³ , ⁴⁴ , ⁴⁵ These differences may reflect epilepsy‐related dysbiosis as well as normal age‐related variation between pediatric patients and adults.

In this study, within the phylum Bacillota, the class Bacilli was significantly more abundant (p < .05), whereas the class Clostridia was significantly less abundant (p < .05) in patients compared with their parents.

At the family level, Streptococcaceae was elevated in patients compared to parents, including the genus Streptococcus (p = .05) (Table S3). Zhou et al. also found Streptococcus to be elevated in pediatric patients with epilepsy compared with healthy controls. ⁴⁴ This increase may likewise be associated with heightened neuroinflammatory responses—particularly in patients with focal epilepsy—as further supported by findings from other inflammatory conditions. ⁴⁴ , ⁴⁶ Other genera, namely Cronobacter, Blautia, Escherichia/Shigella, Colinsella, and Megamonas are reported to be elevated in pediatric patients compared to healthy controls. ²⁹ , ³³ , ⁴⁴

Conversely, patients exhibited lower abundances of several taxa considered beneficial, including genera within the families Lachnospiraceae and Ruminococcaceae. These families comprise well‐known producers of SCFAs, particularly butyrate, which supports gut barrier integrity and exerts anti‐inflammatory effects. ⁴⁷ The depletion of these SCFA‐producing bacteria may also be correlated with the inflammatory status related to epilepsy, as SCFAs are known to play a key role in maintaining gut and immune homeostasis. ⁴⁸ Importantly, evidence from a recent meta‐analysis indicates that the Ruminococcaceae family is significantly depleted in patients with intractable epilepsy compared with healthy controls, which corroborates the results of the present study. ⁴⁹

Interestingly, the genus Faecalibacterium and the family Coriobacteriaceae also trended to lower levels in patients than in parents. As Faecalibacterium is a predominant butyrate producer, its decreased abundance may be associated with inflammation‐related pathways in epilepsy, consistent with previous findings. ⁴⁴ , ⁴⁷ Conversely, the trend toward lower Coriobacteriaceae levels observed in our patients compared with their healthy parents contrasts with the increased abundance of Collinsella within this family reported by Zhou et al. ⁴⁴

Moreover, the abundance of Verrucomicrobia and Akkermansia did not differ between patients and healthy controls, despite higher levels reported in patients with epilepsy in a recent meta‐analysis by Mousavi et al. ⁵⁰

No significant changes in α‐diversity or β‐diversity were evident between T1 and T2 in patients, in accordance with other studies. ²⁹ , ³⁰

Non‐significant changes in Bacillota, Bacteroidota, and Actinomycetota were shown after MedKD, with decreasing trends for Bacteroidota and Actinomycetota and an increase in Bacillota, contrasting with Zhang et al. reporting significantly increased Bacteroidota and decreased Bacillota. ⁵¹ The decrease in Actinomycetota was primarily driven by a reduction in Bifidobacterium spp. (p < .05), as expected in the context of a low carbohydrate, low fiber diet. ⁵²

Several genera increased at T2, notably Clostridium innocuum group, Thomasclavelia (or C. ramosum), Romboutsia, and Ruthenibacterium, many of which are associated with protein and fat metabolism. ⁵³ , ⁵⁴ Notably, the post‐MedKD increase in the Clostridium innocuum group and Thomasclavelia genus—which were already more abundant in patients compared to their parents at baseline—further supports the notion of a compromised immune system in pediatric epilepsy patients. Clostridium innocuum and Thomasclavelia belong to the family Erysipelotrichaceae of Bacillota (or Firmicutes) phylum, which has been linked to inflammation‐related disorders. ⁵⁵ On the contrary, the increase in Extibacter genus is particularly interesting, as some members are involved in bile acid transformation, potentially contributing to the antiseizure effects of the KD via gut–brain axis modulation. ⁵⁶

Some Actinomycetota (or Actinobacteria) species also increased, such as the genus Gordonibacter (family Eggerthellaceae), which is unable to ferment glucose and is associated with fat metabolism. Gordonibacter is also associated with the increased dietary intake of the polyphenols ellagitannins, mostly found in berries and nuts, as part of a Mediterranean‐style diet. Ellagitannins require the activity of Gordonibacter species to undergo reductive and dehydroxylation reactions, leading to the formation of bioactive metabolites known as urolithins, which exhibit anti‐inflammatory and antioxidant properties. ⁵⁷ Unknown Eggerthellaceae, with a relation to plant‐derived polyphenols fermentation, was also increased at T2, potentially related to MedKD nutrients. ⁵⁸ , ⁵⁹

Beneficial butyrate‐producing genera, such as Lachnospira and the Lachnospiraceae ND3007 group, along with Bifidobacterium—a lactate and acetate producer resulting in butyrate production through cross‐feeding interactions—were significantly reduced following the initiation of the MedKD. ⁴⁷ , ⁶⁰ Notably, some of these taxa were already present at lower levels in patients compared to their parents prior to the intervention. This further decline may result from reduced fermentable carbohydrates under the MedKD, leading to lower butyrate production. Veillonella and Haemophilus genera also decreased significantly, suggesting improved inflammatory status linked to the diet's efficacy. To note, the observed alterations in gut microbiota profiles could be associated not only with the adoption of the MedKD but possibly also with the increase in BMI. ⁶¹ , ⁶²

No significant differences in major SCFAs—acetate, propionate, and butyrate—in patients compared to their parents were observed at T1, with the exception of minor SCFAs, such as caproic and heptanoic acid, which were significantly lower in patients. These SCFAs, though less abundant, are involved in lipid metabolism and gut–brain signaling and may influence neurological and immune function. ⁶³ A trend toward higher isocaproic acid levels in patients could indicate altered protein fermentation. ⁶⁴ Although total SCFA balance remained largely similar, differences in these minor metabolites may point to subtle changes in gut–microbiota function relevant to epilepsy pathophysiology. ⁶⁵

In the present study, a decrease in TVFAs was observed following MedKD intervention. Previous studies have reported inconsistent changes in SCFA levels. For example, Gong et al. observed a trend toward increased SCFA levels after 6 months of KD in DRE patients. ²⁹ In contrast, 1 month of a classical KD (4:1 ratio) in DRE patients resulted in a significant reduction of 55% in total SCFAs, particularly acetate, propionate, and butyrate. ⁶⁶ This finding aligns with the results of the present study, which also observed a decrease in TVFAs, although to a lesser extent, with a 19% reduction (p = .055). However, the trend for increased isocaproic acid may reflect increased protein fermentation in response to the diet's macronutrient profile. ⁶⁴ Despite the decrease in butyrate producing bacteria, the levels of key SCFAs remained relatively stable, suggesting that the MedKD did not markedly disrupt core microbial fermentation pathways. To note, the high proportion of olive oil as well as other constituents of the MedKD applied may have contributed to the observed SCFA stability, as olive oil's unsaturated fats and phenolic compounds may potentially influence microbial metabolism and SCFA production to the opposite direction of that of a generally high‐fat diet. ⁹ , ⁶⁷ , ⁶⁸

5. LIMITATIONS

This study has several limitations, including a small sample size, a short time frame and heterogeneous epilepsy etiology. The wide age range and lack of age‐matched controls limited specificity in microbiota analysis. Variability in AEDs use also acted as a confounding factor, as different drugs may differently affect gut microbes. Finally, the small number of participants limited the ability to compare gut microbiota and metabolite alterations between responders and nonresponders following the MedKD.

6. CONCLUSIONS

The findings of this study indicate that patients exhibited a distinct epilepsy‐related gut profile compared to their healthy parents. The MedKD improved adherence and efficacy without major lipid or phylum‐level changes. Microbial shifts reflected reduced carbohydrate fermentation, enhanced fat metabolism, and partial gut balance restoration, likely influenced by high olive oil content. These changes may contribute to seizure control and subtle modulation of SCFA production. Overall, the MedKD is effective and well tolerated in Greek pediatric patients. Future research should examine specific nutrient effects on the gut microbiota to advance precision‐based dietary therapies for epilepsy.

FUNDING INFORMATION

This clinical trial received financial support from Danone Nutricia Greece (Numil Hellas S.A.) to cover expenses related to next‐generation sequencing procedures. Danone Nutricia Greece (Numil Hellas S.A.) had no role in the study design, patient recruitment, data collection, data interpretation, or the decision to submit the manuscript for publication. The investigators maintained full independence in the conduct of the study, analysis and interpretation of the sequencing data, and reporting of the results.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflict of interest.

Test Yourself.

1. Ketogenic diet may

   1. induce gut dysbiosis due to a decrease in butyrate producing bacteria
   2. partially restore bacterial populations in favor of the patient
   3. none of a or b
   4. both a and b
2. The olive oil–based ketogenic diet may have offered a high adherence rate due to

   1. the special components of olive oil
   2. the Mediterranean diet principles are potentially already adopted from the Greek population
   3. an experienced dietitian's guidance
   4. a and c
   5. b and c
3. The stable key SCFA levels after a MedKD for 3 months may be related to

   1. the phenolic compounds of the olive oil contained in the MedKD.
   2. the limited time period of 3 months.
   3. a and b.

Answers may be found in the supporting information .

Supporting information

Data S1.

Data S2.

Data S3.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.