The fecal metabolome and microbiome are altered in dogs with idiopathic epilepsy compared to healthy dogs

Abstract

Idiopathic epilepsy (IE) is the most common chronic neurological disease in dogs, and a natural animal model for human epilepsy types with genetic and unknown etiology. The microbiota-gut-brain axis (MGBA) is a promising target for improving brain health in individuals where brain function is hampered. It’s role in the pathophysiology of epilepsy remains however unclear. We aimed to identify differences in fecal metabolome and microbiome between healthy and dogs with IE. To this purpose, fecal samples of healthy (n = 39) and dogs with IE (n = 49) were metabolically profiled (n = 148 metabolites) and fingerprinted (n = 3690 features) using liquid chromatography coupled to mass spectrometry, and the bacterial phylogeny examined using 16 S rRNA sequencing. Dogs with IE were categorized as drug-resistant (DR) (n = 27) or mild phenotype (MP) (n = 22). In dogs with DR IE compared to healthy, fecal metabolites such as histamine (P = 0.022) and microbiome genera such as Escherichia-Shigella (P = 0.021) increased, associated with a proinflammatory environment. In dogs with MP IE compared to healthy, alterations associated with anti-inflammatory properties, such as increased fecal serotonin (P = 0.034) and Blautia hominis (P = 0.012) were revealed. Overall, a role for the MGBA communication in canine IE was established.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-09919-7.

Introduction

Epilepsy is a chronic non-communicable disease of the brain that is characterized by recurring spontaneous seizures. It affects 50 million people worldwide¹with a detrimental effect on the quality of life of people with the disease as well as their families^2,3. Moreover, one-third of people with epilepsy cannot achieve a life without seizures with the currently available antiseizure medication (ASM)⁴. In 32% of cases, the etiology of human epilepsy remains unknown⁵ highlighting important knowledge gaps. Further exploration of mechanisms involved in epilepsy, such as e.g. the microbiota‒gut‒brain axis (MGBA), is therefore needed.

The MGBA comprises complex bidirectional communication between the gastrointestinal (GI) system and central nervous system (CNS) via neuroanatomical pathways, endocrine, immune and metabolic signaling⁶. The importance and effect of the bidirectional connection between the intestinal microbiome and brain health has been documented for e.g. Parkinson’s disease⁷major depression⁸ and Alzheimer’s disease⁹. As such, the MGBA is an interesting and very promising target for improving brain health in individuals where brain function is hampered, such as those with epilepsy. In epilepsy specifically, research on the MGBA has focused on inflammatory signaling and neuroinflammation¹⁰. Other pathways such as neurotransmitter and amino acid metabolism are however also likely involved in the MGBA communication in epilepsy. These pathways can be studied by applying holistic omics approaches. By concurrently analyzing microbial composition, i.e. through microbiomics^11,12and the resulting metabolic outputs, i.e. through metabolomics, an integrative approach enabling a multi-layered view of the biochemical environment within the GI system can be obtained, revealing functional insights that surpass those achieved through single-omics studies¹³.

Canine idiopathic epilepsy (IE) serves as an established animal model for human epilepsy types with genetic and unknown etiologies¹⁴. Canine and human epilepsy share similar electrophysiological and pharmacological characteristics¹⁴in addition to clinical features such as status epilepticus and behavioral comorbidities¹⁵. Pets moreover often share their living environment with humans, leading to a largely similar exposure to environmental drivers for health and disease, i.e. the exposome^16,17. The nutritional management of pets offers a significant advantage, being that a standardized nutritional background can be easily implemented. Dogs can be fed an identical qualitatively balanced kibble diet that meets their nutritional requirements, thus ruling out nutritional inadequacies. This is particularly important in omics studies, where nutrition is a major and highly variable, environmental determinant¹⁸. As in humans with epilepsy, one third of dogs with IE are inadequately managed with ASM¹⁹indicating the need for novel approaches such as targeting the MGBA. Consequently, advancements in canine epilepsy research have the potential to benefit both veterinary and human medicine.

In this work, we investigated the role of the MGBA in canine IE through integrative intestinal microbiome and metabolome comparison of healthy dogs to dogs with IE using, respectively, 16 S rRNA sequencing and ultra-high performance liquid chromatography coupled to high-resolution mass spectrometry (UHPLC-HRMS). These insights could further clarify the role of MGBA in IE, thereby paving the way for novel management opportunities that could contribute to improved seizure control and quality of life in dogs and humans.

RESULTS

Clinical characteristics of the dogs

The clinical characteristics of the study population were described previously²⁰. Briefly, we enrolled 39 healthy and 49 dogs diagnosed with idiopathic epilepsy (IE), in accordance with the International Veterinary Epilepsy Taskforce guidelines²¹as Tier level I (n = 36) or Tier level II (n = 13). Among the recruited dogs diagnosed with IE, 22 dogs fulfilled the criteria for mild phenotype (MP), and the remaining 27 dogs were categorized as drug-resistant (DR). Both healthy and dogs with IE did not receive antibiotics at least 3 months prior to sampling. No gastro-intestinal (GI) symptoms were recorded in the history of enrolled dogs, except for one dog (DR) with a transient acute episode of diarrhea 21 days prior to sampling, which was addressed by the primary veterinarian with a single dose of maropitant and kaopectate. The most common breeds were Border Collie (n = 17), Crossbred (n = 6), Cane Corso (n = 5) and Golden Retriever (n = 5). Among these 88 dogs, 33 were female (23 castrated), and 55 were male (25 castrated), with a mean age of 4.7 ± 2.2 years and a mean body weight of 26.3 ± 12.8 kg (range 4.4–61.0 kg) at the start of the study. In total, 64 dogs were adults (30/39 healthy; 17/22 MP; 17/27 DR), 15 dogs were seniors (6/39 healthy; 5/22 MP; 4/27 DR), and 9 dogs were geriatric (3/39 healthy; 0/22 MP; 6/27 DR). No significant differences in the abovementioned clinical characteristics were detected (Supplementary Tables S1-S4). The mean body condition score (BCS) was 5.23 ± 0.99, with a significantly higher BCS found in DR IE (P = 0.003), but not in MP dogs, than in healthy dogs. Both healthy and IE dogs received an identical adult maintenance kibble diet for 35 ± 11 days (20–91 days) prior to fecal collection.

In the three months preceding sample collection, dogs with IE experienced a seizure frequency between 0 and 10 seizures per month; the mean seizure frequency (MSF) for DR dogs was 2.9 ± 2.2, and that for MP dogs 0.2 ± 0.3 seizures/month. The median (interquartile range; IQR) time between the last epileptic seizure and fecal sampling was significantly longer (P < 0.001) for MP, i.e., 117 (IQR: 46) days than for DR dogs, i.e., 6 (IQR: 15) days. The age of epileptic seizure onset for dogs with IE was 2.5 ± 1.5 years, whereas no significant difference was observed between MP and DR dogs (Supplementary Tables S1-4). Cluster seizures and status epilepticus were present in 30 dogs, i.e., 61.2% (19 DR and 11 MP), and 13 dogs, i.e., 26.5% (8 DR and 5 MP), respectively. Furthermore, all the IE dogs except 3 (MP) received antiseizure medication (ASM). Among these dogs, 27 dogs received poly- and 19 monotherapy. Phenobarbital, potassium bromide and/or levetiracetam were used in 37 (23 DR and 14 MP), 22 (15 DR and 7 MP) and 11 (7 DR and 4 MP) dogs, respectively. Imepitoin was used in 5 dogs (2 DR and 3 MP), whereas clonazepam and CBD-oil were each used in one DR dog as part of their seizure management. Seizure characteristics and type of ASM were not significantly different between MP and DR dogs.

Targeted metabolite profiling

Among the 148 targeted fecal metabolites that met the inclusion criteria (Supplementary Table S5), six were significantly altered between healthy, DR and MP IE dogs (Fig. 1). Inosine was significantly lower in the feces of DR dogs as compared to healthy (P = 0.027) and MP IE dogs (P = 0.009). Serotonin was significantly higher in the feces of MP dogs as compared to healthy (P = 0.034) and DR dogs (P = 0.012). Both fecal histamine and 1-methylhistamine were significantly higher for DR vs. healthy dogs (P = 0.022 and P = 0.024, respectively). Indole-3-carboxylic acid was significantly lower in the feces of MP (P = 0.036) and DR (P = 0.012) as compared to healthy dogs. Finally, fecal 3-4-dimethoxyphenylacetic acid was significantly lower for MP vs. healthy dogs (P = 0.020).

Fig. 1.

Schematic overview of the significantly altered fecal metabolites in IE compared to healthy dogs. Boxplots represent the iQC normalized peak areas of each metabolite per group. Significant comparisons (P < 0.05) are indicated with ‘*’. His: histidine, Trp: tryptophan, HDC: histidine decarboxylase, PLP: pyridoxal-5-phosphate, HNMT: histamine-N-methyl transferase, DR: drug-resistant (red), MP: mild phenotype (yellow), healthy (green), SD: standard deviation, 3,4 dimethoxyphenylAA: 3,4 dimethoxyphenylacetic acid. This figure was created by the authors using biorender.com.

Although the level of fecal indole was not significantly different between the healthy, MP and DR groups, a significant negative association with MSF was found via linear regression (Table 1). Additionally, the linear regression model revealed three metabolites positively associated with MSF, i.e., 7-ketodeoxycholate, cholic acid and trans-4-hydroxyproline. Finally, a trend towards positive association with MSF was observed for 6 additional metabolites (Table 1).

Table 1.

Results of the generalized linear model “mean seizure frequency ~ metabolite + body condition score + sex + phenobarbital usage + potassium bromide usage + levetiracetam usage”. Metabolites for which P < 0.10 are presented, whereby significant P-values (P < 0.05) are indicated in bold. SD = standard deviation.

Metabolite	Estimate	SD	P
7-Ketodeoxycholate	0.786	0.273	0.007
Indole	−0.792	0.329	0.021
Trans-4-hydroxy-proline	0.812	0.367	0.033
Cholic acid	0.861	0.395	0.035
8-Hydroxyquinoline	0.792	0.396	0.053
Tryptophan	0.659	0.343	0.062
N-Acetyl-methionine	0.645	0.366	0.086
D/β-Alanine	0.694	0.399	0.090
Carnitine	0.523	0.302	0.091
N6,N6,N6-trimethyl-lysine	0.492	0.292	0.099

Fecal metabolic fingerprints

Untargeted analysis generated 3690 features, of which 2220 in the positive and 1470 in the negative ion mode. The discriminative power of the fecal metabolome was examined by building orthogonal partial least squares discriminative analysis (OPLS-DA) models for the different pairwise comparisons, resulting in three OPLS-DA models being compliant with the set validation criteria (Fig. 2), i.e., healthy vs. IE, healthy vs. DR and healthy vs. MP. From the validated OPLS-DA models, 15 features with a variance importance in projection (VIP) score > 1, S-plot correlations |p(corr)| > 0.4, and jackknife confidence intervals not across zero could be retained, thus consistent with good discriminative quality. Of these, 3 features discriminated MP, and 13 features discriminated the DR fecal metabolome from that of healthy dogs. Among these features, one unidentified feature discriminated both MP and DR patients from healthy dogs, and four features could be putatively identified using the Chemspider database (MSI level II²²). Putative 4-hydroxyphenobarbital²³which should indeed only be present in the feces of dogs receiving phenobarbital, 1-beta-hydroxycholic acid²⁴a bile acid, and Boc-L-asparagine (Boc-Asn-OH)²⁵a derivative of the amino acid asparagine, were higher in DR patients vs. healthy controls. Putative 2-(2-carboxyethyl)−4-methyl-5-pentyl-3-furoic acid²⁶a type of heterocyclic fatty acid, was lower in the feces of DR vs. healthy dogs.

Fig. 2.

OPLS-DA score plots of the fecal metabolic fingerprints, with each dot representing an individual dog. Untargeted data were normalized based on internal quality controls (iQC), log transformed, and Pareto scaled prior to plotting. (a) OPLS-DA of healthy vs. IE (b) OPLS-DA of healthy vs. MP and (c) OPLS-DA of healthy vs. DR. The validation parameters of each model are displayed at the bottom of the respective plot. Furthermore, a good permutation plot (n = 100) was obtained for a, b and c. HC: healthy (green); IE: idiopathic epilepsy (blue), i.e., MP and DR; MP: mild phenotype (yellow); DR: drug-resistant (red).

Pathway enrichment analysis (Fig. 3) revealed involvement of vitamin B6 metabolism (P = 0.003), primary bile acid biosynthesis (P = 0.017) and taurine and hypotaurine metabolism (P = 0.035). Within these, eicosapentaenoic acid was putatively identified (MSI level II²²) and increased fecal levels in IE compared to healthy dogs were observed. Within the vitamin B6 pathway, the algorithm putatively identified pyridoxine and pyridoxamine. We could however not confirm this identification using analytical standards, and therefore, this result was discarded. The accompanying chromatograms are displayed in Supplementary fig. S1.

Fig. 3.

Pathway enrichment plots of positively (a) and negatively (b) ionized untargeted metabolic features whereby all matched pathways are presented as circles. The color and size of each circle corresponds to its transformed combined P-value. Large (r) and dark (er) red circles are considered the most perturbed pathways. P-values for the gene set enrichment analysis (GSEA) and mummichog algorithms are displayed on the x- and y-axes, respectively. BA: bile acid; PLP: pyridoxal-5-phosphate, i.e., vitamin B6; UFA: unsaturated fatty acids; cys: cysteine; met: methionine; ala: alanine; asp: aspartate; glu: glutamate. This figure was created by the authors using the functional analysis module in MetaboAnalyst 6.0²⁷.

Fecal Microbiome screening

Alpha diversity was not significantly different between groups (Chao1 P = 0.57, Shannon P = 0.63). Beta diversity indices, visualized in Fig. 4, were significantly different between healthy and MP dogs (Jaccard P = 0.01, unweighted Unifrac P = 0.02) and between healthy and DR dogs (unweighted Unifrac P = 0.04). However, when relative abundance and phylogenetic distance were considered via the weighted UniFrac index (Figure S2), no significant differences remained (healthy vs. DR P = 0.65; healthy vs. MP P = 0.38; MP vs. DR P = 1.00).

Fig. 4.

Principal coordinates plot (PCoA) of the fecal microbiota of healthy dogs (HC), dogs with mild phenotypic (MP) or drug-resistant (DR) idiopathic epilepsy. Each dot represents the fecal microbiome of one dog. The distance between dots represents the difference in microbiome composition between samples, as calculated with the Jaccard (top) and unweighted Unifrac (bottom) metric.

The microbial composition differed between IE and healthy dogs, when evaluating relative abundancies (Fig. 5). Relative abundancies for all phylogenetic levels are provided in Supplementary table S6, and sequencing data is publicly available at the NCBI sequence read archive (PRJNA1252412). At the genus level, an increase in fecal Clostridium sensu stricto 1 (P = 0.001; L2FC = 3.26) and Escherichia-Shigella (P = 0.021; L2FC = 3.55) was detected in DR compared to healthy dogs. Moreover, a decrease in fecal Succinivibrio (P = 0.026; L2FC = −4.41) and Phascolarcobacterium (P = 0.042; L2FC = −2.39) was detected in MP compared to healthy dogs. When the amplicon sequence variant (ASV) level was examined, no significant differences between DR and healthy dogs remained. However, in the feces of MP dogs, one bacterial species was increased, i.e., Blautia hominis (P = 0.012; L2FC = −1.44), and three species were decreased, i.e., Phascolarcobacterium succinatutens (P = 0.012; L2FC = 5.11), Segatella copri DSM 18205 (i.e., Prevotella 9, P = 0.016; L2FC = 3.70) and Succinivibrio spp. (P = 0.016; L2FC = 4.47), compared to healthy dogs. For the latter, no putative identification at the species level was possible.

Fig. 5.

Differentially abundant genera (a) and amplicon sequence variants (ASVs) (b) in dogs with idiopathic epilepsy compared to healthy dogs. Boxplots show the relative abundances of the differentially abundant genera for drug-resistant (DR) and mild phenotype (MP) dogs compared with healthy controls (HC). Each dot represents the respective relative abundance in the feces of one dog. This figure was created by the authors using the R DeSeq2²⁸ and ggplot2 (v.3.5.1) packages.

Correlations between metabolites and microbes

Spearman correlation analysis revealed 677 significantly correlated metabolite‒ASV pairs, whereby 3 pairs were associated with a | ρ | > 0.4. Benzoic acid was positively correlated with both ASV 161, i.e., Clostridium hiranonis (P < 0.001; ρ = 0.42), and ASV 210, i.e., Bacteroides (P < 0.001; ρ = 0.45). Creatine was negatively correlated with ASV 66, i.e., Segatella copri DSM 18205 (P < 0.001; ρ = − 0.40). Moreover, for differentially abundant ASVs, an additional 61 metabolite-ASV pairs were identified with 0.4 > | ρ | > 0.2 (Supplementary table S7).

DISCUSSION

Our study revealed alterations in fecal metabolome and microbial composition in IE compared with healthy dogs, while the nutritional background was standardized by providing the same adult maintenance diet to all dogs for minimal 20 days before fecal collection. These findings support a role for the MGBA in the pathophysiology of canine IE, and provide promising insights into the underlying mechanisms involved in this bidirectional communication. In addition to the biological interpretation of the findings, it is essential to acknowledge the limitations and strengths associated with the employed methodologies and study design.

Metabolic alterations in feces of dogs with IE related to inflammation

The levels of both histamine and 1-methylhistamine were higher in the feces of DR dogs than in those of healthy dogs. In the GI system, histamine is typically derived from L-histidine by histidine decarboxylase (HDC) and is degraded by histamine-N-methyl transferase (HNMT) in the cytosol or by membrane-bound HNMT to 1-methylhistamine³⁷. Histamine can either be produced by immune cells in the periphery of the host, such as mast cells or dendritic cells³⁸ or by certain types of bacteria³⁹. Under normal physiological conditions, mast cells⁴⁰ and dendritic cells⁴¹ can cross the blood-brain barrier (BBB), linking host-produced histamine in the periphery to histamine in the brain. In mammals, HDC is an enzyme dependent on pyridoxal-5-phosphate (PLP), i.e. the active form of vitamin B6⁴², whereas bacterial HDC can use either pyruvoyl or PLP as a coenzyme³⁹. Interestingly, the authors recently described a decreased PLP plasma concentration in the same study population of dogs with IE compared with healthy dogs, despite a standardized nutritional background²⁰.The observed increase in fecal histamine/1-methylhistamine could be linked to the decrease of plasma PLP in dogs with IE. However, this decrease in PLP was observed in both MP and DR dogs, whereas the levels of fecal histamine and 1-methylhistamine were significantly increased in DR dogs only. The increase in Clostridium sensu stricto 1 and Escherichia-Shigella in dogs with DR might thus play a role in this, as these bacteria have the genetic potential to use pyruvoyl-dependent HDC³⁹. Moreover, histamine sensing in Escherichia coli, a member of the Escherichia-Shigella genus, can increase the bacterial catabolism of short-chain fatty acids (SCFAs), thereby interfering with the regulatory role of SCFAs in the intestine⁴³ and promoting a proinflammatory environment. On the other hand, the increased fecal histamine could indicate decreased GI barrier integrity, leading to leakage of histamine. Hereby, the timing between last seizure and sampling could explain the difference between DR and MP, as acute effects on GI barrier integrity are expected, similar to what is seen after stroke in humans⁴⁴. Regardless of the source, intestinal histamine contributes to peripheral inflammation via histamine H₂ receptors (H₂Rs), leading to BBB disruption and the recruitment of other immune cells, ultimately connecting the innate immune response in the intestines to the brain⁴⁵. In the central nervous system however, a complex interaction between histamine, neural excitability and epilepsy exists, whereby the effect on seizure propagation in different animal models is dependent on the location and type of receptor (H_1,2,3 or H₄R)⁴⁶. Several animal studies have shown that histamine decreases in different brain regions during focal and generalized seizures. Therefore, a protective effect of high brain histamine levels was suspected, although it is unclear how brain histamine levels flux during the different stages of the disease epilepsy⁴⁶. The increased fecal histamine observed in DR dogs in our study is however unlikely to be related to central histamine, and an indirect effect on the CNS through peripheral inflammation is hypothesized.

Inosine, which was decreased in the feces of DR compared to MP and healthy dogs, is a degradation product of adenosine. Adenosine metabolism increases under stressful conditions such as inflammation, producing more inosine. Inosine in turn acts as a signaling molecule that can modulate the immune system⁴⁷. An increase in blood purines, i.e., inosine, hypoxanthine and xanthine, was correlated with seizure severity and neurodegeneration in mouse models⁴⁸. In people with temporal lobe epilepsy, increased baseline (> 24 h following a seizure event) blood purine levels were found compared to those in controls⁴⁸. Moreover, anti-epileptic properties for inosine have been described in different tonic‒clonic seizure animal models, including mice⁴⁹ and zebrafish⁵⁰. Conversely, pro-epileptic properties for uric acid and inosine were found in a rat model of ‘absence seizures’⁵¹. The decrease in fecal inosine in DR dogs observed in our study suggests that more inosine is degraded by the GI microbiota or less inosine is excreted. A negative correlation was found between Cl. sensu stricto 1 and hypoxanthine (Supplementary table S7), i.e. an inosine metabolite, which would be consistent with an increased microbial breakdown in DR dogs. On the other hand, if less inosine is excreted, more inosine is potentially metabolized to hypoxanthine, xanthine and uric acid or transported to the brain in dogs with DR IE. This may be linked to the perturbation of the taurine and hypotaurine pathways in IE, detected via pathway enrichment analysis, as taurine can attenuate xanthine oxidase⁵²the enzyme required to transform xanthine into uric acid⁵³. Uric acid as such has been linked previously to cytotoxic brain injury and oxidative stress⁵⁴aspects that might contribute to DR in dogs.

Metabolic alterations in feces of dogs with IE related to Tryptophan metabolism

An increase in fecal serotonin was observed in MP compared to both DR and healthy dogs. Peripheral serotonin cannot cross the BBB under normal physiological conditions⁵⁵. However, over 90% of the body’s serotonin is produced in the intestines, mainly by enterochromaffin (EC) cells, overflowing to the GI lumen and blood circulation⁵⁶. Thus, the fecal serotonin detected in the current study corresponds to the leftover EC-produced serotonin pool. The EC serotonin pool targets primary afferent neurons in the GI tract, signaling nausea and discomfort from the GI tract to the brain, and regulating peristalsis and secretion in the GI tract⁵⁷. Murine studies have shown that intestinal bacteria play a regulatory role in body serotonin levels⁵⁸. More specifically, E. coli can have the genetic ability to produce serotonin, in addition to PLP, tryptophan and indole⁵⁹and might therefore be involved in alterations in these metabolites. Previously, anticonvulsive properties and reduced mortality have been associated with higher CNS serotonin levels in humans and mice⁶⁰. However, it is unclear whether the EC serotonin pool could have similar effects mediated by signaling from the GI system to the brain. It has been demonstrated that this signaling pathway involves the afferent fibers of the vagal nerve⁶¹which is a common target for neuromodulation in epilepsy in both humans⁶² and dogs⁶³.

Indole, a bacterial degradation product of tryptophan, was negatively associated with the MSF in IE dogs and tryptophan as such tended to be positively associated with MSF. These associations could indicate greater bacterial degradation of tryptophan in dogs with a lower MSF. Thereby, indole enhances the intestinal epithelial barrier function⁶⁴which might reduce the leakage of inflammatory mediators to the peripheral circulation, and ultimately the CNS. Indole-3-carboxylic acid, another indole metabolite derived from tryptophan, was significantly lower in the feces of IE (both MP and DR) compared to healthy dogs. Simultaneously, higher serotonin levels were detected in MP (i.e. dogs with a lower MSF) than in DR dogs. Moreover, a negative correlation between tryptophan and Segetella copri and Succinivibrio spp., both decreased in MP dogs, was found, indicating that both spp. may be involved in the microbial turnover of tryptophan. Ultimately, we hypothesize that dogs with a lower MSF have greater turnover of tryptophan to serotonin and indole⁴³both exerting potential positive effects, i.e. stimulation of the vagal nerve and enhancement of the epithelial barrier, respectively.

Fecal metabolic alterations differ between dogs with MP and DR IE

Compared to those in healthy dogs, the metabolic alterations were different for dogs with DR and MP IE, revealing an interaction between seizure frequency, severity, time since last epileptic seizure or response to ASM and the fecal metabolome/microbiome. This may be cause or consequence, i.e. the fecal metabolome and microbiome in MP IE dogs may have a protective effect, leading to a lower MSF and/or a better response to ASM. Previously, a study revealed that mice receiving fecal microbiota from an epileptic mouse showed an increase in seizure susceptibility compared to recipients of healthy microbiota⁶⁵. Moreover, the presence of systemic inflammatory diseases has the capacity to aggravate epileptogenesis⁶⁶. Therefore, the seizure characteristics in our study might be caused by the different GI metabolic environments. On the other hand, it is possible that peripheral alterations induced by the condition, such as inflammation, are less pronounced in MP compared to DR IE as a consequence of the milder phenotype or longer time between fecal collection and last epileptic seizure. Epilepsy as such can increase the BBB permeability⁶⁷potentially attributing to systemic alterations beyond CNS involvement. Hereby, the potential confounding effect of the time between the last epileptic seizure and sampling should indeed be considered. This time interval was significantly lower in DR dogs, with a median time interval of 6 days compared to a median time interval of 117 days in MP dogs. A shorter interval may have resulted in more acute alterations caused by epileptic seizures compared to longer intervals in MP dogs; where no acute effects are presumed.

Microbial alterations in feces of dogs with IE

Compared with those of healthy dogs, the feces of dogs with MP and DR presented significantly altered differential abundances of multiple bacterial genera. These alterations were different for each group; however, no significant differences were observed when the fecal microbiome compositions of DR and MP dogs were compared directly. As a consequence of the characteristics of 16 S rRNA sequencing, elaborated below under ‘5. Strengths and limitations’, caution is needed when comparing our findings to literature. However, a recent study comparing drug-naïve IE and healthy dogs revealed similar findings, i.e. lower relative abundancies for Prevotella spp. and Phascolarctobacterium together with higher relative abundancies in Escherichia-Shigella and Clostridium sensu stricto 1 in dogs with IE⁶⁸. Moreover, all alterations found in our study, both for MP and DR dogs, were also observed previously in dogs with idiopathic inflammatory bowel disease (IBD) in comparison to healthy dogs⁶⁹although no accompanying clinical GI symptoms were observed in our study. While alterations in microbial composition differ between dogs and humans⁷⁰a similar IBD pathophysiology comprising an immune-mediated basis, influenced by genetic and environmental factors, has been proposed⁷¹. Further similarities with IBD could be substantiated when evaluating the metabolic fingerprint, whereby our study putatively showed altered primary bile acid biosynthesis in dogs with IE. Whereas in dogs with IBD, one of the major metabolic alterations included a decrease in secondary bile acids, i.e., deconjugated primary bile acids formed by the GI microbiome⁷². According to a human population-based study, IBD increases the risk for epilepsy, with an adjusted hazard ratio of 1.30⁷³. Although there are, to the best of our knowledge, no studies linking IE and IBD in dogs, our fecal metabolome and microbiome findings support a bidirectional link between the CNS and an inflammatory GI environment.

Compared to those in healthy dogs, the abundances of the genera Clostridium sensu stricto 1 and Escherichia-Shigella were greater in dogs with DR IE. The latter is in line with microbiome alterations in humans, whereby an increase was observed in adults with poststroke epilepsy⁷⁴in infants with epilepsy and comorbid diarrhea⁷⁵in children with focal epilepsy in the pretreatment phase⁷⁶as well as in an adult epilepsy cohort⁷⁷. Moreover, a shift toward reduced carbohydrate metabolism, linked to an increase in the Escherichia genus, was observed in children with severe epilepsy on a ketogenic diet, raising concerns about the functional microbial alterations caused by this type of diet⁷⁸. However, in IE dogs receiving a specific type of ketogenic diet, i.e., medium chain triglycerides (MCTs), no significant alterations in the genus Escherichia have been noted^79,80. Interestingly, in the latter of these MCT dog studies, lower levels of the Blautia genus were detected in DR vs. MP dogs at baseline⁸⁰whereas in our study, Blautia hominis was increased in MP compared to healthy dogs. In the MCT dog study by Pilla and colleagues, a decrease in Blautia spp. compared with the baseline was observed in both the placebo and MCT-diet groups⁷⁹. However, no conclusion could be drawn regarding the reason for this decrease. In humans, Blautia spp. are generally considered to have anti-inflammatory properties^81,82. A generalized conclusion of effects at the genus level should be avoided but considered at the species or even strain level⁸³. In the context of the gut‒brain axis, Blautia hominis was more abundant in an elderly human population not experiencing postoperative delirium (n = 20) compared to elderly people experiencing postoperative delirium (n = 20)⁸⁴suggesting a positive effect in the context of MGBA communication.

Surprisingly, at the ASV phylogenetic level, only significantly different ASVs were found between MP and healthy dogs. Segetella copri DSM 18205, whose abundance was decreased in MP dog feces, has been shown to produce proinflammatory cytokines and promote a Th-17-mediated immune response in vitro^85,86. This further substantiates our hypothesis that MP dogs exhibit fewer proinflammatory characteristics in their intestinal environment than DR dogs. Dogs with MP IE also presented a lower abundance of Phascolarctobacterium succinatutens and Succinivibrio spp., both of which are involved in propionate production in dogs⁸⁷. Propionate, as such, is important in the regulation of hepatic lipid metabolism and satiety in the host⁸⁸. It could be hypothesized that the ASV alterations in MP dogs are therefore potentially related to a positive response to ASM.

Study strengths and limitations

The strengths of our study are the simultaneous analysis of both fecal metabolomics and microbiomics, providing an in-depth characterization of changes in the GI system of IE compared to healthy dogs. To the best of our knowledge, multi-omics integration at this level has been performed only once in subjects with naturally occurring epilepsy, i.e., children with DR epilepsy²⁹. Our study is also the first to include an MP group, as well as standardized nutrition, hence excluding interfering dietary effects. The latter should not be overlooked, because mice³⁰ and human³¹ studies have indicated reversible, yet pronounced effects of dietary perturbations on GI microbiome, reaching a steady state at 2 to 3.5 days following nutritional interventions. Therefore, we hypothesize that 20 days of standardized nutrition is adequate in reducing interindividual variation caused by nutrition. In the microbiome, functional redundancy is typically present, emphasizing the importance of the role for specific bacteria in addition to their phylogenetic composition³². Our study attempted to include functional information by combining microbiome and metabolome data³³revealing significant alterations related to inflammation and tryptophan metabolism. However, the specific interactions between GI microbiome and metabolome could not be fully elucidated, due to complex interaction between host, GI metabolome and microbiome, together with analytical considerations.

Metagenomic sequencing, including both 16 S sRNA sequencing (as applied here) and whole genome sequencing (WGS) were associated with poor interlaboratory reproducibility³³. One study even reported higher interlaboratory than interindividual variability³⁴complicating comparison between studies. However, technical variation within the current study was minimized by utilizing the same protocol for all runs³⁵ and providing a mock community as positive control in each run³⁴. Nonetheless, the hypothesis generating role for 16 S sequencing should be emphasized, highlighting the need for future targeted approaches, like qPCR³⁶. Conversely, the applied metabolomics method was analytically validated⁹¹ ensuring the reliability of the detected metabolic alterations.

Since all dogs with IE except 3 received ASM at the time of sampling, this is considered as an important confounder. Future research should therefore aim to include drug-naïve dogs and evaluate the effects of ASM on the fecal metabolome and microbiome. Furthermore, the time between the last seizure and sampling is another important confounder, especially in the comparison between MP and DR dogs. Future studies should therefore consider multiple sampling timepoints related to the last seizure in dogs with IE to discriminate acute seizure effects from chronic alterations.

Conclusion

Dogs with IE presented an altered fecal metabolic fingerprint and profile, and differences in microbiome composition compared to healthy dogs (on a standardized nutritional background). Overall, the feces of dogs with DR IE presented alterations suggesting a proinflammatory GI environment, i.e. an increase in histamine and 1-methylhistamine, and a decrease in inosine, as well as elevated Escherichia-Shigella and Clostridium sensu stricto 1. In contrast, the feces of dogs with MP IE revealed increased serotonin and inosine compared to healthy and DR dogs. In addition, Blautia hominis was increased, whereas the genera Succinivibrio and Phascolarctobacterium were decreased. The noted changes have previously been linked to positive GI health effects, including anti-inflammatory properties and enhanced epithelial barrier function.

Our study suggests a role for the MGBA in the pathophysiology of canine idiopathic epilepsy, potentially interfering with response to ASM and/or seizure frequency. However, it is unclear whether the GI microbiome and metabolic alterations are cause or consequence of epilepsy. Our multi-omics approach also unraveled novel metabolic pathways of interest, including histamine and tryptophan metabolism, which could be further exploited in future targeted research and intervention studies.

Materials and methods

Study design and subjects

The study protocol (Fig. 6), approved by the Ethical Committee of Veterinary Medicine and Bioscience Engineering, Ghent University (EC2020-091), included three groups of client-owned dogs: healthy dogs, MP IE dogs, and DR IE dogs. The inclusion criteria for all dogs included unremarkable anamnesis (or consistent with IE), unremarkable general physical and neurological exams, and no abnormalities on blood or urine tests. The BCS was recorded on a 9-point scale, with 4–5 points considered ideal⁸⁹.

Fig. 6.

Study protocol. Following recruitment, all enrolled dogs were transitioned to a standardized adult maintenance diet for a minimum duration of 20 days prior to fecal sample collection. Following fecal sampling, DNA was extracted for 16 S sRNA sequencing and metabolites were extracted for UHPLC-HRMS metabolomics analysis. HC: Healthy controls; MP: mild phenotype idiopathic epilepsy; DR: drug resistant idiopathic epilepsy; UHPLC-HRMS: ultra-high performance liquid chromatography coupled to high-resolution mass spectrometry. This figure was created by the authors using biorender.com.

For dogs with IE, alterations in liver enzymes (ALT, ALP) and electrolytes (K, Na, Cl) without clinical relevance and related to ASM were retained. No antibiotics were allowed three months prior to sample collection. IE diagnosis followed the International Veterinary Epilepsy Taskforce guidelines, i.e. Tier I (based on signalment, medical history, video evaluation of a seizure and routine blood examination) or II (Tier I + bile acids, MRI and cerebrospinal fluid evaluation)²¹. An epileptic seizure diary for three months prior to the start of the study was needed, allowing the classification of dogs. Within the MP group dogs with ≤ 1 seizure/3 months and no status epilepticus nor cluster seizures, and dogs with a good response to ASM, i.e. >50% seizure reduction post-ASM adjustment, were included. Whereas dogs were categorized as DR if there was failure to achieve > 50% seizure reduction despite ≥ 2 ASMs for ≥ 2 months⁹⁰.

All dogs received an identical adult maintenance diet and treats at the start of the study, per FEDIAF 2019 guidelines based on maintenance energy requirements (MER). The key nutrient composition of the diet is provided in supplementary table S8. The owners were instructed to adhere strictly to the provided diet for at least 20 days before fecal collection. Freshly voided fecal samples were collected within 15 min, frozen, and stored for a maximum of 24 h at home by the owners until cooled transport. After arrival at the research facility, samples were aliquoted for DNA extraction and 16 S rRNA sequencing (stored at −20 °C), as well as for metabolome analysis (stored at −80 °C). Fecal aliquots for metabolome analysis were subjected to 72 h of lyophilization to eliminate microbial activity and facilitate homogenization. Plasma samples from the same study were collected and analyzed, of. which results were published previously²⁰.

Metabolomics analysis

Generic extraction, followed by UHPLC-HRMS analysis of the fecal metabolome, was performed via a validated extraction protocol and analysis method⁹¹. In-house standard mixtures of 371 metabolites (Supplementary Table S5) were used to validate the operational conditions, together with quality control (QC) samples prepared from a representative pool of fecal sample extracts (n = 50) and a negative control sample, i.e., a blank sample prepared following the extraction protocol. Standard mixtures were injected at the beginning and end of the sequence. QC samples and a negative control sample were injected at the beginning, following every 10 biological samples and at the end of the sequence. Biological samples were analyzed in a randomized order⁷².

Microbiome analysis

Bacterial DNA for microbiome analysis was extracted via the QIAamp^® PowerFecal Pro DNA extraction kit. One sample could not be included in the microbiome analysis owing to technical issues during DNA extraction (DR; dog 95), resulting in 87 analyzed fecal samples (39 healthy, 26 DR and 22 MP). Positive controls (n = 4; ZymoBIOMICS^® Microbial Community Standard) were included to assess the quality of sample preprocessing and sequencing. Bacterial barcoding was conducted through a 2-step amplification process utilizing the primers S-D-Bact-0341-b-S-17 (5′-CCTACGGGNGGCWGCAG-3′) and S-D-Bact-0785-a-A-21 (5′-GACTACHVGGGTATCTAATCC-3′), which amplify the V3-V4 region of the 16 S rRNA gene as previously described⁹². The final barcoded libraries were sequenced on two different runs via Illumina MiSeq v3 technology (2 × 300 bp, paired-end) by Macrogen.

Data processing and statistical analysis

Clinical parameters

Clinical parameters are described as proportions (categorical data) or means ± standard deviations. Furthermore, dogs are categorized as adult, senior or geriatric dogs based on their age and body weight, since small dog breeds are known to live longer than large dog breeds⁹³. Categorical data were compared using a Pearson Chi-square test. Numerical data were analyzed using one-way ANOVA followed by a post-hoc Tukey’s test for normally distributed data or Kruskal-Wallis rank sum followed by Dunn’s test for non-normally distributed data. IE characteristics included the type of epileptic seizure (generalized tonic clonic seizure (gtcs) or focal seizure)⁹⁴presence of cluster seizures or status epilepticus in the medical history, age at seizure onset, time between sampling and last seizure, and MSF in the 3 months preceding sample collection. Within the IE dogs, numerical data were compared between MP and DR using an independent two-sided T-test or Mann-Whitney U test. For all statistical tests a P-value < 0.05 was considered significant. Details regarding the clinical parameters of the dogs included in the study can be found in Supplementary Table S1−4 and Verdoodt et al., 2025 (Epilepsia)²⁰ where the same population was used to elucidate plasma metabolome alterations in dogs with IE compared with healthy dogs.

Metabolomics

Both targeted metabolites and untargeted features were normalized via internal QCs (iQCs). After iQC normalization, only metabolites with a coefficient of variance (CV) < 30% in the QCs were retained for further analysis.

Targeted metabolomics

For the in-house available metabolite standards, peak areas were obtained via manual integration via Xcalibur^™ 4.1 (Thermo Fisher Scientific, USA). Only metabolites with a signal-to-noise ratio > 3 in 90% of the samples were considered. Identification was achieved based on accurate mass (m/z value, considering both the molecular ion and C₁₃-isotope) and retention time relative to that of an external standard (level I identification according to the Metabolomics Standards Initiative (MSI))²². Further data processing was executed via Excel (Microsoft, USA) and R (v 4.2.3)⁹⁵.

First, fecal metabolite levels were semi-quantitatively compared between groups, i.e., healthy, IE, IE MP or IE DR, via a univariate approach⁹⁶. Three-group comparisons (healthy vs. MP vs. DR) were performed via one-way ANOVA for normally distributed metabolites or a Kruskal‒Wallis rank sum test for non-normally distributed metabolites. The significance (P < 0.05) of each pairwise comparison, i.e. healthy vs. DR, healthy vs. MP and DR vs. MP, was then evaluated using a post hoc Tukey (following ANOVA) or Dunn (following Kruskal-Wallis) test.

Second, a generalized linear model was built that included only samples from IE dogs to evaluate the influence of known metabolite levels on the MSF. Within this model, age, sex, BCS and the use (yes vs. no) of the most common ASM, i.e., phenobarbital, potassium bromide and levetiracetam, were added as confounders. The results of the model include an estimate, i.e., an estimated change in the MSF for a one-unit change in each respective metabolite, assuming that all confounders are held constant, a standard error associated with this estimate and a P_lin value, considered significant if P_lin < 0.05 or a trend if 0.05 < P_lin < 0.10. Positive estimates indicate higher fecal levels of the metabolite with a higher MSF.

Untargeted metabolomics

Untargeted data preprocessing was performed with Compound Discoverer (CD) 3.3 (Thermo Fisher Scientific, USA), combining positive and negative ions. The detected features were characterized by the m/z value (peak intensity threshold of 500.000 a.u., mass tolerance of 5 ppm), retention time (RT; maximum RT-shift of 0.4 min) and peak intensity (minimal signal-to-noise ratio of 3). Data preprocessing included log transformation and Pareto scaling. Further data processing was executed in Simca 17.1 (Umetrics AB, Sweden) and MetaboAnalyst 6.0²⁷. In Simca, an unsupervised multivariate PCA-X model (including the QC samples) was built for exploration of inherent sample variance and QC clustering. The sample data (excluding the QC samples) were then modeled via supervised OPLS-DA, whereby 4 pairwise comparisons were made: healthy vs. IE, healthy vs. DR, healthy vs. MP and DR vs. MP. The model parameters R²(Y) for fit, Q²(Y) for predictivity (both > 0.5), cross-validated analysis of variance (CV-ANOVA, P value < 0.05) and permutation testing (n = 100) were assessed to evaluate model validity. The discriminative quality of features was investigated on the basis of variance importance in projection (VIP) scores > 1, S-plot correlations |p(corr)| > 0.4, and jackknife confidence intervals not across zero. In parallel, pathway enrichment analysis via the Mummichog and GSEA algorithms in MetaboAnalyst 6.0 was performed for each ionization method separately (without additional filtering steps in the MetaboAnalyst environment), and the fecal metabolome of IE (DR and MP) dogs was compared to that of healthy dogs. Putative identification was pursued whenever possible by matching measured m/z values (< 5 ppm difference) to theoretical m/z values and retention times in the Chemspider or in-house database (level 2 identification according to MSI)²².

Microbiomics

The demultiplexing of the amplicon dataset and the deletion of the barcodes were carried out by the sequencing provider. All further statistical analyses were performed in R (v 4.2.3)⁹⁵. The raw sequence reads were trimmed, quality filtered and dereplicated via the dada2 package (v 1.24.0)⁹⁷. First, an initial ASV table was constructed, followed by removal of the chimeras via the removeBimeraDenovo function. Second, the taxonomy was assigned via dada2’s naïve Bayesian classifier method on the basis of the Silva database (v 1.38)⁹⁸. A phylogenetic tree was then constructed via the DECIPHER (v 2.24.0) algorithm⁹⁹after which a neighbor-joining tree was constructed via phangorn (v 2.10.0)¹⁰⁰. The resulting phylogenetic tree and ASV table were loaded into phyloseq (v 1.38.0)¹⁰¹whereby a minimal sequencing depth of 10 000 reads was achieved. Within the phyloseq package, alpha (Chao1 and Shannon) and beta (Jaccard, unweighted and weighted UniFrac) diversity indices were calculated and statistically compared between groups via PERMANOVA. Furthermore, ASVs whose total abundance was less than 0.01% and whose prevalence was less than 50% within a group were discarded before further analysis. Significantly (P < 0.05) differentially abundant bacterial taxa were identified at different phylogenetic levels by applying DESeq2²⁸ to the resulting compositional data, considering the confounders sex, BCS and age category (adult, senior or geriatric). The taxonomy of significantly differentially abundant ASVs was checked by comparing the differentially abundant ASV sequences to multiple taxonomic databases via the Basic Local Alignment Search Tool (BLAST)¹⁰²whereby a minimal identical percentage of 98% was considered for identification.

Multi-omics integration

The correlation between targeted metabolites and the microbiome was evaluated via Spearman’s correlation test¹⁰³resulting in a correlation coefficient ρ and a false discovery rate (FDR)-corrected P value. A metabolite‒ASV correlation was considered significant when P < 0.05. Of these, only correlations where | ρ | > 0.4 were retained for biological interpretation. Additionally, significant pairs with differentially abundant ASVs and 0.4 > | ρ | > 0.2 were reported as supplementary information.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

F.V. contributed to the conception and design of the study; the acquisition, analysis and interpretation of the data; and the drafting of the manuscript. L.Y.H. and S.F.M.B. contributed to the conception and design of the study; acquisition, analysis and interpretation of the data; and critical revision of the manuscript. L.V. and E.G. contributed to the analysis and interpretation of the data and critical revision of the manuscript. L.V.H., F.V.I., J.M. and M.H. contributed to the interpretation of the data and critical revision of the manuscript. All the authors have read and approved the final version of the manuscript.

Funding

Fien Verdoodt (1S71421N) is supported as an SB PhD fellow by the Research Foundation–Flanders (FWO). Lieselot Y. Hemeryck (1297623N) is supported by the Research Foundation - Flanders (FWO).

Data availability

Data on the analytical standards and detailed clinical characteristics of the enrolled dogs are provided in the supplementary information. The raw metabolomics data are available via https://www.ebi.ac.uk/metabolights/MTBLS10915. Raw sequencing data are available at the NCBI sequence read archive (PRJNA1252412).

Declarations

Competing interests

F. Verdoodt is currently working on a doctoral research project, including the current study, regarding the role of the gastrointestinal microbiome and nutrition in canine IE, which is financially supported by Nestlé Purina Petcare Europe.M. Hesta is a member of the Advisory Board of Nestlé Purina Petcare, and has been paid for several consulting services by a variety of pet food companies.E. Goossens has no financial or personal relationships with other people or organizations that could inappropriately influence or bias the content of the paper.F. Van Immerseel has no financial or personal relationships with other people or organizations that could inappropriately influence or bias the content of the paper.J. Molina is employed by Nestlé Purina Petcare Europe.L. Van Ham has no financial or personal relationships with other people or organizations that could inappropriately influence or bias the content of the paper.L. Vanhaecke has no financial or personal relationships with other people or organizations that could inappropriately influence or bias the content of the paper.L.Y. Hemeryck has no financial or personal relationships with other people or organizations that could inappropriately influence or bias the content of the paper.S.F.M. Bhatti has no financial or personal relationships with other people or organizations that could inappropriately influence or bias the content of the paper.

Ethical declaration

This research was conducted in compliance with European legislation on animal experimentation (EU directive 2010/63/EU) and the ARRIVE guidelines. The study protocol was approved by the ethical committee of the Faculty of Veterinary Medicine and Bioscience Engineering, Ghent University (EC2020-091). Written informed consent was obtained from the owners to participate in the study, and collect samples for research purposes. In adherence with the European privacy regulations (EU Regulation 2016/679), the collected samples cannot be traced back to the dog owners based on the published information.