Comparison of clinical efficacy and gut microbiota characteristics in children with ASD treated with fecal microbiota transplantation and ketogenic diet Authors: Yu Wang, Lihong Wang, Zhifeng Cai, Lianhu Yu, Yanshuo Guo, Linlin Zhang, Mengna Zhu, Zhiyue Liu, YongHeng Zhao, Lili Liu, Aihua Cao Published: 2026 ABSTRACT ObjectiveAutism Spectrum Disorder (ASD) is a neurodevelopmental disorder characterized by impairments in social communication and interaction, along with restricted, repetitive patterns of behavior. It is often accompanied by gastrointestinal dysfunction and gut microbiota dysbiosis. Fecal Microbiota Transplantation (FMT) and the Ketogenic Diet (KD) are interventions targeting the gut microbiota for ASD.Methods30 participants were diagnosed with ASD according to DSM-5 and ADOS-2. ASD core symptoms were evaluated with CARS and ABC. Gut microbiota composition was analyzed by shotgun metagenomic sequencing.ResultsBoth groups demonstrated significant improvements in core symptoms. In the FMT group, the mean CARS score significantly decreased from 34.87 to 33.53 (p < 0.01); in the KD group, it declined from 35.13 to 33 (p < 0.01). The mean ABC score reduced from 79.93 to 69.33 (p = 0.064) in the FMT group and from 63.07 to 42.73 (p < 0.01) in the KD group. Following the intervention, no statistically significant changes were observed in α-diversity or β-diversity within either group. LEfSe analysis revealed distinct post-intervention microbial signatures: FMT significantly enriched butyrate-producing taxa (Wujia chipingensis, Eubacterium sp. MSJ-33, and Butyrivibrio crossotus), while KD elevated Blautia massiliensis and decreased propionate metabolism -associated taxa (Veillonella sp. S12025-13 and Veillonella nakazawae). KEGG enrichment analysis revealed that KD enriched propionate metabolism (Fold enrichment = 3.747, q = 0.010) and aromatic compound degradation (Fold enrichment = 3.591, q = 0.010).ConclusionsBoth interventions significantly improved clinical symptoms among children with ASD, potentially through distinct patterns of gut microbiota modulation.Clinical trials numberNCT06348433 (03/21/2024).Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s12888-026-08547-3. Introduction Autism spectrum disorder (ASD) is a neurodevelopmental disorder characterized by communication impairments and repetitive, restrictive behaviors [1]. The prevalence of ASD has shown a significant upward trend, increasing from 1 in 150 individuals in 2000 to 1 in 31 by 2022 [2, 3]. Changes in diagnostic criteria and increased awareness among parents and healthcare professionals have contributed to the rising prevalence. Evidence suggests that, in addition to genetic susceptibility, environmental factors may also contribute to the development of ASD [4]. Environmental factors associated with ASD are multifaceted and may contribute to metabolic abnormalities, oxidative stress, and immune dysregulation [5, 6]. In addition, many children with ASD exhibit gastrointestinal symptoms, such as diarrhea and gastroesophageal reflux, which are linked to impaired epithelial barrier function, immune dysregulation, and marked gastrointestinal inflammation [7, 8]. Previous studies further suggest that gut microbiota interact with environmental factors and influence gastrointestinal manifestation through immune and metabolic mechanisms [9, 10]. Taken together, these findings highlight the gut microbiota as a promising target for developing the therapeutic strategies in ASD. Among the emerging microbiota-targeted interventions for ASD, fecal microbiota transplantation (FMT) and the ketogenic diet (KD) have received particular attention. FMT involves transferring standardized gut microbiota from healthy donors to patients to restore the intestinal microenvironment [11, 12]. By contrast, the KD modulates gut microbiota through a high-fat, low-carbohydrate nutritional regimen and has more moderate effects [13]. Kang et al. demonstrated that FMT can ameliorate both gastrointestinal and core symptoms in ASD patients, yielding long-term benefits [14]. Following FMT, the gut microbiota and intestinal environment resembled those of the donor, with diversity indices exceeding baseline levels. Importantly, the abundance of beneficial bacteria, including Bifidobacterium and Prevotella, was increased and maintained up to two years [15]. Subsequent studies further corroborated that modulation of gut microbiota can induce positive shifts in intestinal metabolites (e.g., short-chain fatty acids) and serum neurotransmitters (e.g., dopamine, 5-HT), thereby improving gut barrier integrity and mitigating inflammation [16, 17]. Similar to FMT, KD also showed significant efficacy in improving core ASD symptoms [18, 19]. Evidence indicates that KD can induce rapid alterations in the gut microbiota [20], While this phenomenon has been demonstrated in epilepsy, its role in ASD remains comparatively underexplored. In a clinical study of seven children with ASD, KD intervention significantly increased in α-diversity of the gut microbiota across the family, genus, and species levels [21]. Complementary animal experiments revealed that KD can significantly increase the abundance of Akkermansia and Blautia in ASD model mice, while simultaneously decreasing levels of pro-inflammatory factors and oxidative stress markers in peripheral blood and brain tissue [22, 23]. These findings suggest that KD may reduce systemic inflammation and central nervous system oxidative stress by remodeling gut microbiota composition, and ultimately improving ASD-related behavioral symptoms. Collectively, both FMT and KD demonstrate potential for modulating gut microbiota composition and alleviating ASD symptoms. Key uncertainties persist regarding their clinical interchangeability, complementary utility, and differential efficacy across distinct ASD symptom domains. Elucidating their mechanisms and applicability boundaries is therefore crucial for selecting optimal individualized therapeutic regimens. Current ASD treatment primarily relies on behavioral interventions and rehabilitation training, as no effective pharmacological treatments are available for its core symptoms [24]. Increasing evidence implicates the gut microbiota as a potential etiological factor, offering new therapeutic insights for ASD [25, 26]. This study primarily investigates the safety of FMT and the KD, and concurrently evaluates their effects on core symptoms of ASD and gut microbiota composition. By directly comparing these two interventions, we aim to establish a scientific foundation for individualized treatment strategies, assisting clinicians in selecting the most suitable microbiota-targeted therapeutic approach for each child with ASD. Methods Inclusion criteria and exclusion criteria This clinical trial was conducted in accordance with the 1964 Helsinki Declaration and its later amendments and ethical standards that are consistent with Good Clinical Practice. The ethical approval was given by the Ethics Committee of Shandong University Qilu Hospital. This trial was registered with ClinicalTrials.gov (NCT06348433). A total of 30 children (26 males and 4 females, aged 19 to 97 months) were recruited from Shandong University Qilu Hospital. All enrolled participants were diagnosed with ASD using the Autism diagnostic observation schedule, second edition (ADOS-2) and Diagnostic and statistical manual of mental disorders, fifth Edition (DSM-5). The participants were required to have not taken probiotics, antibiotics, or immunosuppressants for at least one month before enrollment in the study. The study’s exclusion criteria encompassed cases of secondary autism attributable to identifiable factors, including metabolic disorders or genetic diseases, fever with a body temperature surpassing 37.3 °C, carnitine deficiency, beta-oxidation defects, inability to maintain adequate nutrition, serious gastrointestinal conditions necessitating urgent treatment (such as intestinal obstruction, hemorrhage, or perforation), significant immunodeficiency, a history of severe allergies, profound malnutrition or being underweight, as well as children receiving specialized dietary therapy. As both FMT and KD are emerging interventions for ASD children without standardized treatment protocols, and considering the young age of the participants and the difficulty in recruitment, the sample size was determined based on both feasibility and preliminary power estimation. Using preliminary data, we estimated a mean reduction of 1.0 point in CARS scores with a standard deviation of 1.25, corresponding to an effect size of 0.8. Power analysis was performed using G*Power 3.1. With a two-tailed α of 0.05 and 80% power, the minimum required sample size was estimated to be 15 patients per group for detecting within-group change using a paired t-test. Therefore, 15 patients were included in each group to evaluate the safety and preliminary efficacy of these interventions. Figure 1 represents the research procedure.Fig. 1CONSORT diagram CONSORT diagram Intervention The intervention protocols for FMT and the KD are summarized below (Fig. 2). The present study was a parallel controlled clinical trial with a non-random allocation ratio approximately 1:1, determined by the child’s ability to swallow capsules and parental preference. Participants capable of oral capsule intake underwent FMT group, whereas those unable to do so or whose parents refused FMT were assigned to the KD group.Fig. 2The intervention protocols for FMT and the KD The intervention protocols for FMT and the KD FMT intervention Fifteen children with ASD participated in a 9-week study involving 5 weeks of FMT therapy followed by a 4-week observation period. Prior to the initiation of FMT, all ASD participants underwent training to facilitate capsule swallowing. In preparation for the FMT treatment, they adhered to a residue-free semi-liquid diet for three days to cleanse the gastrointestinal tract and enhance gut motility, thereby promoting the colonization of gut microbiota. Moreover, each participant was administered GOLYTELY (a polyethylene glycol solution) one day prior to the FMT treatment to ensure effective intestinal evacuation. FMT capsules were administered orally during the first week and again in the fifth week. After completing the five-week treatment phase, participants were closely monitored for an additional four weeks. Throughout the FMT regimen, all participants maintained their routine rehabilitation training. Efficacy evaluations were conducted at week 9 (T2) post-intervention initiation, corresponding to the end of the 4-week post-treatment observation period, in order to capture the short-term sustained effects after completion of the FMT course. KD intervention Fifteen children with ASD underwent intervention using a Modified Atkins Diet (MAD) protocol, with macronutrient energy contribution ratios set at fat: protein: carbohydrate = 60%:30%:10%. Prior to intervention, the medical team formulated individualized meal plans for each participant and provided systematic training for primary caregivers. Training content encompassed the mechanism of MAD, identification and management of potential adverse reactions (e.g., hypoglycemia and ketoacidosis), and home-based dietary management strategies, with particular emphasis on carbohydrate counting, calorie requirement estimation, protein adequacy, and micronutrient supplementation protocols. A dynamic adjustment strategy was implemented during the protocol execution: in the initial phase, fasting or 2-hour postprandial blood glucose and ketone concentrations were monitored three times daily (target ranges: approximately 3.9–5.0 mmol/L for glucose; 2.0–5.0 mmol/L for ketones). Certified nurses demonstrated operational procedures and trained caregivers in measurement techniques. Based on metabolic adaptation status, monitoring frequency was progressively reduced from daily to once every three days, then to weekly assessments. Concurrently, ≥1 weekly medical follow-up was maintained, during which registered dietitians dynamically optimized dietary plans according to weight trajectories and ketosis status. All subjects maintained their pre-existing rehabilitation regimens throughout the intervention period. Efficacy evaluations were conducted at week 12 (T2) post-intervention initiation, allowing sufficient time for metabolic adaptation, stabilization of ketosis, and establishment of dietary adherence under the MAD protocol. FMT donor screening and capsule preparation All fecal microbiota capsules used in this study batch were derived from a single unrelated male donor aged 11 years. As the donor was a minor, donor-related procedures were completed with guardian involvement and documented informed consent/authorization. Donor eligibility was assessed using a structured health questionnaire covering general health status, medication history, gastrointestinal symptoms, allergy history, infectious disease history, family history, bowel habits, and recent travel/exposure history. Standardized laboratory screening included routine blood testing, liver function assessment, serum infectious disease screening, stool pathogen testing, multidrug-resistant organism screening, and SARS-CoV-2 testing. Stool pathogen screening included PCR detection of Clostridioides difficile toxin B and coronavirus, quantitative PCR for norovirus, culture for common enteric bacterial pathogens, EIA-based testing for selected enteric pathogens, and microscopic examination for parasites and ova. Serum screening included HIV-1/2 antigen/antibody, hepatitis A/B/C-related markers, syphilis, and Strongyloides IgG antibody testing. Fecal microbiota capsules were prepared by a professional microbiology laboratory under anaerobic conditions, including filtration, centrifugation, lyophilization, and encapsulation, followed by product quality-control testing. Assessments and diagnosis of autism symptoms This study implemented a multidimensional assessment system comprising the ADOS-2, Childhood autism rating scale (CARS) and Autism behavior checklist (ABC). All children in both the FMT and KD groups underwent the complete battery of all three assessments at baseline (T0) and at their group-specific endpoint (T2). The ADOS-2 serves as the gold standard for clinical ASD diagnosis [27], utilizing a semi-structured, observation-based methodology. Tailoring the module to each participant’s age and ability, we used calibrated severity scores (CSS) to neutralize inter-module variability and directly compare Social Affect (SA) and Restricted and Repetitive Behaviors (RRB) profiles. The CARS [28]—a pivotal diagnostic behavioral instrument—incorporates 15 domains rated on a 4-point scale (1 = normal to 4 = severely abnormal; total range 15–60). The ABC [29] documents 57 behavioral characteristics in children with autism, encompassing five domains: Sensory, Relating, Body and Object Use, Language, and Social and Self-help. Total scores range from 0 to 158, with scores ≥ 67 suggesting ASD. The scores of all the above scales were positively correlated with the severity of ASD symptoms. Fecal sample collection and metagenomic sequencing In accordance with the International Human Microbiome Standards (IHMS), a total of 48 fecal samples were prospectively collected and immediately stored at −80 °C after collection. Baseline samples (T0) were obtained from 12 participants in each group (FMT and KD), and matched endpoint samples (T2) were collected at the corresponding follow-up time points. The gut microbiota metagenomic sequencing protocol employed standardized procedures: Genomic DNA was extracted from intestinal samples using Hangzhou Guhe Nucleic Acid Extraction Kit (GHFDE100, Hangzhou Guhe Biotechnology, Hangzhou, China), followed by concentration measurement via Qubit 2.0 fluorometer (Q32866, Life Technologies, Carlsbad, CA, USA) and integrity verification through 1% agarose gel electrophoresis (TAE buffer system, Super GelRed staining, Biotium, Fremont, CA, USA, 20 min). Subsequent library construction involved enzymatic fragmentation generating 350–450 bp DNA fragments, end repair/dA-tailing, and size selection using AMPure XP beads (A63881, Beckman Coulter, Brea, CA, USA) through dual-step purification—first eliminating > 500 bp fragments then capturing target fragments (300–500 bp) from supernatant. Sequencing adapters were ligated with T4 DNA ligase at 20 °C for 30 min followed by magnetic bead purification to remove unbound adapters. PCR amplification (14 cycles: 98 °C 10s → 60 °C 20s → 72 °C 20s) using Yi Sheng 2×Hieff Canace PCR Master Mix (Yeasen Biotechnology, Shanghai, China) preceded final bead-based size selection to obtain 300–500 bp libraries. Quantified libraries meeting Qubit standards were subjected to adapter ligation efficiency assessment via SYBR Green qPCR, culminating in equimolar pooling and dilution to target concentration for 2 × 150 bp paired-end sequencing on the Illumina NovaSeq 6000 platform (Illumina, San Diego, CA, USA) using a 300-cycle S4 Reagent Kit (20012866, Illumina). Statistical analysis Clinical scale data were analyzed using SPSS 25.0 software (IBM Corp.), with continuous variables described as mean values. Normally distributed data were analyzed using the t-test, while non-normally distributed variables were assessed with the Wilcoxon test. Demographic group differences were evaluated via the Kruskal-Wallis test. For gut microbiota analysis, raw reads were quality-controlled using FastQC software before and after filtering. Low-quality reads were removed according to the following criteria: reads with an average quality score ≤ Q20 per 4 bp window were discarded; reads shorter than 50 bp were excluded; adapter sequences, ambiguous bases (N), and host-derived reads were also removed. For taxonomic profiling, clean metagenomic sequences were annotated to species level using Kraken2 (version 2.1.3), with relative abundances quantified via Bracken (v2.8.0). Subsequent bioinformatic and statistical analyses were performed using GraphPad Prism (v9.0) and R (v4.0.3). Age effects on microbiota profiles were evaluated with a linear mixed-effects model incorporating age as a covariate. Alpha diversity was calculated using the Chao1 richness index, Shannon diversity index, and Simpson diversity index. Beta diversity was visualized through Principal Coordinate Analysis (PCoA) and Non-metric Multidimensional Scaling (NMDS) based on Bray-Curtis dissimilarity. Differentially abundant bacterial taxa were identified using Linear Discriminant Analysis Effect Size (LEfSe) with an LDA score threshold of 2.0, and results displayed in boxplots. For functional metagenomics, reference genomes from NCBI were translated into protein sequences and aligned to the HUMAnN2 UniRef90 database. KEGG orthology annotations were assigned using eggNOG-mapper (v2.0.1). Differential abundance of microbial genes was detected via DESeq2 (v1.44.0)‘s negative binomial Wald test, with significance thresholds set at adjusted p < 0.05 (Benjamini-Hochberg correction, BH correction) and absolute log2 fold-change > 1. Enriched KEGG pathways were analyzed using the hypergeometric test in the clusterProfiler R package (v4.12.6); pathways with nominal enrichment significance (p < 0.05) underwent BH correction to control false discovery rate (FDR ≤ 0.05). All correlation analyses employed Spearman’s rank correlation due to non-normal data distribution. Differences with a p-value of less than 0.05 were considered significant. Results Subject characteristics All ASD participants enrolled in this trial were comparable in most demographic characteristics at baseline (Table 1), including body mass indexes (BMIs), mode of delivery and maternal age during pregnancy. However, in this trial, KD group children were younger than those in the FMT group at baseline, likely due to younger children’s difficulty in swallowing capsules (p = 0.001). The trial also evaluated the supplementation of folic acid and multivitamin during pregnancy, and no significant differences were found between the FMT group and KD group.Table 1Clinical characteristics of ASD participantsCategoryFMT group (n = 15)KD group (n = 15)p valueAge (months), (mean ± SD)68.07 ± 15.3646.73 ± 17.690.001Gender0.224Female03Male1512BMI (kg/m2), (mean ± SD)17.46 ± 2.7917.46 ± 5.130.590Mode of delivery1.000Vaginal delivery77Cesarean section88Maternal age (years), (mean ± SD)30.20 ± 4.7430.13 ± 4.390.968Folic acid supplementation0.139Periconceptional48Early pregnancy116None01Multivitamin supplementation0.520Periconceptional52Early pregnancy67None46 Table 1Clinical characteristics of ASD participantsCategoryFMT group (n = 15)KD group (n = 15)p valueAge (months), (mean ± SD)68.07 ± 15.3646.73 ± 17.690.001Gender0.224Female03Male1512BMI (kg/m2), (mean ± SD)17.46 ± 2.7917.46 ± 5.130.590Mode of delivery1.000Vaginal delivery77Cesarean section88Maternal age (years), (mean ± SD)30.20 ± 4.7430.13 ± 4.390.968Folic acid supplementation0.139Periconceptional48Early pregnancy116None01Multivitamin supplementation0.520Periconceptional52Early pregnancy67None46 Clinical characteristics of ASD participants In the KD group, 10 of 25 participants discontinued the intervention, corresponding to a discontinuation rate of 40%. Discontinuation was mainly related to poor dietary adherence and difficulty maintaining the regimen, especially in ASD children who had marked food selectivity and poorer cooperation. Baseline comparisons showed no statistically significant differences in demographic characteristics or baseline scale scores between completers and dropouts (Supplementary Table 1). Clinical scale assessment of FMT versus KD Significant improvement in core ASD symptoms was observed following both FMT and KD interventions. The baseline characteristics of the enrolled ASD children were assessed using the ADOS-2 scale, which consists of two subdomains: Social affect (SA) and Restricted and repetitive behaviors (RRB). Prior to the intervention, no significant differences in SA (p = 0.941), RRB (p = 0.501) and total ADOS scores (p = 0.964) were observed between the two groups (Fig. 3a). In the FMT group, the average CARS scores declined significantly from 34.87 to 33.53 (p = 0.002), while in the KD group, the scores decreased from 35.13 to 33 (p < 0.001, Fig. 3b). The KD group exhibited a significant decrease in ABC scores from 63.07 to 42.73 (p = 0.002), whereas the FMT group showed a reduction from 79.93 to 69.33 (p = 0.064, Fig. 3c).Fig. 3Therapeutic efficacy comparison between FMT and KD interventions in children with ASD. (a) Baseline ASD symptom severity assessed by ADOS-2, where ADOS represents the CSS for overall ASD symptomatology, SA indicates CSS for the social affect domain, and RRB denotes CSS for the Restricted and repetitive behaviors domain; (b) Change in CARS total scores; (c) Change in ABC total scores; (d) Differential improvement in CARS subdomains; and (e) Differential improvement in ABC subdomains Therapeutic efficacy comparison between FMT and KD interventions in children with ASD. (a) Baseline ASD symptom severity assessed by ADOS-2, where ADOS represents the CSS for overall ASD symptomatology, SA indicates CSS for the social affect domain, and RRB denotes CSS for the Restricted and repetitive behaviors domain; (b) Change in CARS total scores; (c) Change in ABC total scores; (d) Differential improvement in CARS subdomains; and (e) Differential improvement in ABC subdomains Compared with the KD group, the FMT group demonstrated significantly greater improvement in CARS scores, specifically in the domains of relation, imitation, emotional response, and object use. The KD group demonstrated greater improvement in ABC subscale scores, specifically in sensory, relating, and language, while the FMT group also showed significant improvement in the sensory domain (Fig. 3d, e, Supplementary Tables 2–5). Moreover, this trial also refined the analysis of blood cells and liver and kidney function to clarify the safety of FMT and KD intervention. The results showed that there were no abnormal indicators and no statistical significance before and after FMT and KD intervention (Supplementary Tables 6–7). A total of 5 adverse events were reported among the 25 children who received KD intervention, including hypoglycemia (1 case), vomiting (2 cases), diarrhea (1 case) and abdominal distension (1 case). A total of 4 adverse events were reported among the 20 children who received FMT intervention, including constipation (2 cases), diarrhea (1 case) and anorexia (1 case). Most adverse events were gastrointestinal or transient in nature (Supplementary Table 8). Taxonomic shifts in gut microbiota: FMT vs KD To investigate whether clinical improvements were mediated by the gut microbiota, we conducted metagenomic sequencing of fecal samples from 12 FMT-treated and 12 KD-treated ASD children at pre- and post-intervention timepoints. Linear mixed-effects modeling with age as a covariate revealed that age had no significant effect in this clinical trial (Fig. 4a). Both interventions increased alpha diversity indices (Chao1, Shannon, and Simpson), though without significant differences between interventions (Fig. 4b). PCoA and NMDS analysis revealed no significant beta diversity separation between baseline and post-treatment samples in either group (Fig. 4c, d). To further elucidate distinct microbial community differences between FMT and KD groups, LEfSe analysis identified 5 FMT-associated and 3 KD-associated microbial taxa (LDA > 2, p < 0.05, Fig. 4e). The bacteria Wujia chipingensis (LDA score = 4.23, p = 0.008), Eubacterium sp. MSJ-33 (LDA score = 2.97, p = 0.011), and Butyrivibrio crossotus (LDA score = 2.53, p = 0.010), which were significantly enriched following FMT intervention, are known for producing short-chain fatty acids (SCFAs), especially butyrate (Fig. 4f). In the KD intervention group, Veillonella sp. S12025-13 (LDA score = 2.37, p = 0.027) and Veillonella nakazawae (LDA score = 2.30, p = 0.032) showed a decrease. These taxa are involved in lactate and propionate metabolism, potentially impacting neurological functions (Fig. 4f). Spearman correlation analysis revealed significant associations between alterations in specific bacterial species and improvements in core ASD symptoms (Fig. 4g). Within the FMT intervention group, increased abundance of the butyrate-producing Eubacterium sp. MSJ-33 was negatively correlated with the “General Impressions” subscale of the ABC scale (r = −0.405, p = 0.005). Similarly, elevated Butyrivibrio crossotus levels showed a marginally significant negative correlation with the “object use” subscale (r = −0.282, p = 0.058). In the KD intervention group, diminished abundances of Veillonella species (sp. S12025-13 and nakazawae) correlated positively with ABC total scores (r = 0.404, p = 0.004; r = 0.387, p = 0.007) and multiple subdomains—including sensory, body and object use, language, and vsual response (p < 0.05)—indicating their decline aligned with clinical improvement. The increased Blautia massiliensis after KD exhibited a significant negative correlation with the “sensory” subscale (r = −0.309, p = 0.033).Fig. 4Gut microbiota changes following FMT and KD interventions and their correlations with clinical symptoms. (a) Linear mixed-effects model assessing the influence of age on clinical outcomes; (b) Alpha diversity indices (Chao1, Shannon, Simpson) before and after FMT or KD intervention; (c-d) Beta diversity analysis using PCoA and NMDS between pre- and post-treatment samples within either group; (e-f) LEfSe analysis identified 5 FMT-enriched and 3 KD-enriched microbial taxa (LDA > 2, p < 0.05); (g) Spearman correlation analysis between specific bacterial abundances and improvements in ABC scale scores Gut microbiota changes following FMT and KD interventions and their correlations with clinical symptoms. (a) Linear mixed-effects model assessing the influence of age on clinical outcomes; (b) Alpha diversity indices (Chao1, Shannon, Simpson) before and after FMT or KD intervention; (c-d) Beta diversity analysis using PCoA and NMDS between pre- and post-treatment samples within either group; (e-f) LEfSe analysis identified 5 FMT-enriched and 3 KD-enriched microbial taxa (LDA > 2, p < 0.05); (g) Spearman correlation analysis between specific bacterial abundances and improvements in ABC scale scores Functional profiling and correlation analyses of the gut microbiota after FMT and KD intervention Functional metagenomic analysis revealed 25 significantly differential genes pre- versus post-intervention in the FMT group (FDR q < 0.05, |log2FC|≥1; Fig. 5a), though no KEGG pathways were significantly enriched (q > 0.05). In contrast, the KD group exhibited 262 differential genes (FDR q < 0.05, |log2FC|≥1; Fig. 5b) with significant enrichment in propionate metabolism (Fold enrichment = 3.747, q = 0.010) and aromatic compound degradation (Fold enrichment = 3.591, q = 0.010) (Fig. 5c). The propionate pathway involved 9 downregulated genes (e.g., pduW, pdl), collectively indicating suppressed propionate synthesis. Top 20 most significantly altered genes identified 10 photosynthesis-related genes (e.g., apcE, cpcG, pecC) downregulated in FMT. Among these, apcE, cpcC, cpcG, cpeD, and cpeE showed significant negative correlations with the social responsiveness subscale (r = −0.328, p < 0.05) and positive correlations with object use (r = 0.315, p < 0.05) (Fig. 5d). In the KD cohort, 10 upregulated functional genes (e.g., TC. GBP, ttrS, fixJ) demonstrated no significant clinical associations (Fig. 5d). Finally, correlation analysis identified a negative association between Eubacterium sp. MSJ-33 and E2.6.1.14 (r = −0.285, p = 0.049), suggesting butyrate-producing bacteria may optimize precursor utilization for butyrate synthesis by suppressing amino acid metabolism (Fig. 5e).Fig. 5Functional metagenomic signatures following interventions. (a) Differentially expressed genes in FMT group (FMT_T0 vs FMT_T2). (b) Differentially expressed genes in KD group (KD_T0 vs KD_T2). Blue/red dots denote significantly down/upregulated genes (padj<0.05, |log2FC|≥1). Gray points indicate non-significant changes. (c) Sankey-bubble plot of KEGG pathway enrichment following KD. X-axis: Fold enrichment, bubble size: gene count per pathway, color scale: q value. (d) Heatmap of Spearman correlations between top 20 genes and behavioral domains (ABC/CARS scores). (e) Heatmap of Spearman correlations between top 20 genes and microbial species Functional metagenomic signatures following interventions. (a) Differentially expressed genes in FMT group (FMT_T0 vs FMT_T2). (b) Differentially expressed genes in KD group (KD_T0 vs KD_T2). Blue/red dots denote significantly down/upregulated genes (padj<0.05, |log2FC|≥1). Gray points indicate non-significant changes. (c) Sankey-bubble plot of KEGG pathway enrichment following KD. X-axis: Fold enrichment, bubble size: gene count per pathway, color scale: q value. (d) Heatmap of Spearman correlations between top 20 genes and behavioral domains (ABC/CARS scores). (e) Heatmap of Spearman correlations between top 20 genes and microbial species Discussion Numerous previous studies have documented gut microbiota dysbiosis in ASD individuals [30]. The gut microbiota may participate in ASD pathogenesis through multiple pathways, including metabolic, immune, and neural mechanisms [31, 32]. This study found that both FMT and KD intervention methods regulated the intestinal microbiota and significantly improved the core symptoms of ASD. Specifically, the KD group showed more significant improvements in sensory, relating, and language on the ABC scale, while the FMT group demonstrated more prominent improvements in relation, imitation, emotional response, and object use on the CARS scale. It is worth noting that there are differences between the two assessment methods. The ABC scale is more subjective in assessment, while the CARS scale is more objective. At the practical operation level, KD may encounter the problem of operational complexity, and young children may have difficulty swallowing capsules when receiving FMT treatment. As both of these therapies have their own advantages and disadvantages, in practical applications, the most suitable intervention method should be flexibly selected based on the age, symptom characteristics and family support conditions of children with ASD. Based on the significant improvement in clinical symptoms, in-depth analyses of taxonomy and function profiles were conducted using gut microbiota metagenomic data. No statistically significant differences in alpha- or beta-diversity were observed between the two groups either before or after the interventions. This suggests that the overall gut microbiota structure remained relatively stable during the intervention period, potentially attributable to the effects of specific microbial taxa. In this trial, FMT significantly increased the abundance of Wujia chipingensis, Eubacterium sp. MSJ-33, and Butyrivibrio crossotus – taxa which previous reports suggest may be closely associated with butyrate production [33–35]. In animal models, high abundance of butyrate-producing bacteria and butyrate markedly improved intestinal permeability and reduced associated inflammatory responses [36, 37]. Furthermore, butyrate significantly improves blood-brain barrier (BBB) integrity and mitigates neuroinflammation, thereby exerting neuroprotective effects [38]. Consistent with prior research, modulating the gut microbiota via FMT can ameliorate ASD-core symptoms and gut dysbiosis, possibly by enhancing the relative abundance of butyrate-producing species and subsequently increasing butyrate levels [39, 40]. Specifically, butyrate may alleviate symptoms by enhancing intestinal barrier function, restoring BBB integrity, and suppressing neuroinflammation via pathways such as NF-κB [39, 41]. The KD intervention exhibited a distinct microbial modulation pattern compared to FMT: Propionate-producing Veillonella species (V. sp. S12025-13 and V. nakazawae) significantly decreased in abundance, while Blautia massiliensis increased. Consistent with previous studies, a KD alters the abundance of Blautia [22, 42]. Metagenomic analysis revealed significant enrichment of the propanoate metabolism pathway, with coordinated downregulation of 9 core genes (pduW, mmsA, pduD), indicating suppressed microbial conversion of lactate to propionate. Supporting evidence from rodent models demonstrates that exogenous propionic acid (PPA) injection induces ASD-like behaviors, which exhibits neurotoxicity and can directly act on the central nervous system [43, 44]. Furthermore, it compromises BBB integrity, activates the immune system, and triggers neuroinflammatory pathways via PTEN/Akt, MAPK/ERK, and mTOR/Gskβ signaling mechanisms [43, 45, 46]. Studies have also demonstrated that propionic acid disrupts neurotransmitter transmission in the central nervous system, primarily by reducing levels of serotonin and dopamine [47]. Thus, KD may improve core ASD symptoms by reducing neurotoxic PPA exposure through inhibition of gut microbial propionate synthesis. To our knowledge, this is the first study to compare FMT and KD in children with ASD from both clinical and gut microbiota perspectives. Our results show that both interventions were associated with clinical improvement, together with changes in microbial composition and functional pathways, including those related to SCFA metabolism. However, because SCFA levels were not measured directly, the possible roles of butyrate, propionate, and other SCFAs remain speculative. The present findings therefore support an association between microbiota-related changes and clinical improvement, but do not establish an SCFA-mediated mechanism. However, this study has several limitations. First, the relatively small sample size may limit the statistical power and generalizability of the findings to the broader ASD population. Second, the intervention and follow-up durations were short, precluding assessment of long-term safety and sustained efficacy. Third, treatment allocation was not randomized and instead depended on the child’s ability to swallow capsules and parental preference. This pragmatic approach may have introduced selection bias and made the two groups less directly comparable. A further concern is that the KD group was younger at baseline, which may be relevant to both clinical outcomes and gut microbiota composition. Fourth, although significant shifts in microbial taxa and functional genes related to SCFA metabolism were observed, SCFA levels were not directly measured in fecal samples. Therefore, interpretations regarding butyrate- and propionate-mediated mechanisms remain speculative, and the current findings should be understood as associations between microbial changes and clinical improvement rather than evidence of causal metabolic pathways. Future studies incorporating targeted metabolomic profiling will be necessary to validate these hypotheses. A further limitation of this study is the high discontinuation rate in the KD group. This likely reflects the difficulty of maintaining a restrictive ketogenic diet in young children with ASD, especially those with marked food selectivity, repetitive behaviors, and poor cooperation. Because the efficacy analysis was not performed on an intention-to-treat basis, the reported KD results may overestimate the treatment effect and should be interpreted with caution. Finally, the FMT and KD groups were assessed at different post-intervention time points because the two interventions differed in treatment course and expected timing of response. Although these schedules were chosen for clinical reasons, the difference in assessment windows makes it more difficult to compare changes over time and reduces the strength of direct between-group comparison. In conclusion, this exploratory study provides preliminary evidence that both FMT and KD are associated with significant improvements in core autism symptoms and induce distinct alterations in the gut microbiota and its predicted metabolic functions in children with ASD. The KD, in particular, may represent a viable microbiota-targeting intervention strategy, especially for children who cannot receive capsule-based FMT. The interventions appeared safe within the context of this trial. However, given the limitations of sample size and the inferential nature of the functional analysis, these findings must be interpreted as preliminary. We emphasize the potential of targeting the gut-brain axis in ASD and highlight the need for larger-scale, randomized controlled trials that incorporate longitudinal multi-omics data (including metabolomics) to confirm efficacy, establish causality, and elucidate the precise mechanisms involved. Electronic supplementary material Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1