The Microbiota–Gut–Brain Axis in Autism: Associations, Causal Inference, and Interventions—A Narrative Review

ASD represents a highly heterogeneous neurodevelopmental condition characterized by persistent deficits in social communication and interaction, alongside restricted and repetitive behaviors, interests, or activities [1,2]. Its etiology is multifactorial: genetic susceptibility interacts with diverse environmental and biological influences, including immune–inflammatory pathways [3,4] and oxidative stress [5]. No single pathophysiological mechanism fully accounts for the marked phenotypic variability [6,7,8]. Epidemiologic surveillance over the past decade indicates a global rise in ASD prevalence, underscoring its public health significance [9]. Current interventions primarily alleviate symptoms and enhance functioning, while disease-modifying therapies remain unavailable [10,11]. Accordingly, identifying modifiable biological and environmental pathways that contribute to ASD-related phenotypes represents a critical research priority [12,13].

The MGBA describes a bidirectional communication network between the gut and the central nervous system (CNS), mediated by intestinal microbiota [14,15]. This network operates through multiple interacting routes [16]: (1) neural—primarily via the vagus nerve (VN), enteric nervous system (ENS), and sympathetic nervous system (SNS); (2) immune—through innate and adaptive immune cells and cytokine signaling; (3) endocrine—via the hypothalamic–pituitary–adrenal (HPA) axis; and (4) metabolic—through microbially derived metabolites such as short-chain fatty acids (SCFAs) and tryptophan derivatives that can modulate glial activity and neural circuits. This framework offers a conceptual model for exploring how gut physiology may influence brain function and behavior. Nevertheless, current findings are largely correlational, providing mechanistic hypotheses rather than demonstrating causality [17].

Multiple studies indicate that gastrointestinal (GI) symptoms (e.g., constipation, diarrhea, abdominal pain, and bloating) are more prevalent among individuals with ASD [18,19,20]. Reported prevalence varies across studies due to differences in assessment tools, age groups, recruitment settings, and medication use, but meta-analyses estimate that roughly one-third of individuals with ASD are affected—significantly higher than in the general pediatric population [21]. The presence and severity of GI symptoms frequently correlate with greater social communication deficits, restricted and repetitive behaviors, anxiety, and sleep disturbances [22]. However, these associations remain correlational, and causal relationships are not yet established. This pattern highlights the MGBA as a potentially informative framework for investigating the biological underpinnings of ASD's clinical heterogeneity [23].

Integrity of the intestinal barrier is essential for maintaining host–microbiota symbiosis and homeostasis [24]. When dysbiosis or external factors such as diet compromise this barrier, pathogen-associated molecular patterns, including lipopolysaccharide (LPS), can translocate across the mucosa into circulation, activating innate immune pathways such as the NLRP3 inflammasome and inducing interleukin-1β (IL-1β) and interleukin-18 (IL-18) release [25,26,27]. Systemic inflammation may subsequently disrupt BBB integrity through cytokine signaling and loss of endothelial tight junctions, thereby promoting neuroinflammation. Microglial activation and polarization are consistently reported in ASD models and human studies [28,29,30,31]. Recent evidence also implicates central NLRP3 activation in BBB dysfunction. In ASD, in vivo imaging and histological analyses reveal abnormal microglial activation [32]. Nonetheless, most evidence remains correlational and insufficient to infer causality [26,33,34].

Because the enteric (ENS) and sympathetic (SNS) nervous systems are less extensively studied, this section focuses on the VN. Among MGBA pathways, the VN provides the most rapid anatomical and physiological link between the gut and the CNS [35]. As the tenth cranial nerve and the main afferent–efferent branch of the parasympathetic system, the VN descends from the brainstem to the thoracoabdominal cavity, extensively innervating visceral organs, including the gastrointestinal tract [36]. Approximately 80% of its fibers are afferent, establishing a predominantly bottom-up flow of information from the gut to central nodes such as the nucleus tractus solitarius [37,38,39]. Peripheral VN terminals detect luminal and mucosal signals—including microbial metabolites, gut-derived hormones, and inflammatory mediators—and relay them to the brainstem–limbic–cortical network, modulating emotional and cognitive processing [35,37,40,41]. Classic animal studies demonstrate that Bifidobacterium longum NCC3001 induces anxiolytic-like effects in mice with chronic colitis, effects abolished after vagotomy, underscoring the VN's role in microbiota-to-brain signaling [42,43]. The VN also regulates peripheral immunity via the cholinergic anti-inflammatory reflex and the sympathetic–splenic pathway, reinforcing its function as an integrative hub within the MGBA [44].

The endocrine system connects stress and metabolic states to the MGBA through the HPA axis [45,46]. Stress-induced cortisol elevations upregulate the rate-limiting enzymes tryptophan 2,3-dioxygenase (TDO) and indoleamine 2,3-dioxygenase (IDO), shifting tryptophan metabolism toward the kynurenine pathway and reducing serotonin synthesis [47]. Kynurenine metabolites, in turn, modulate neuroimmune signaling and affective behaviors. Evidence from human and animal studies supports these mechanisms, though effect sizes and population variability remain to be clarified [48,49].

MGBA signaling operates through parallel, interconnected neural (vagal and enteric), immune (NLRP3–cytokine), endocrine (HPA), and metabolic (SCFAs and tryptophan–kynurenine) pathways. For instance, peripheral LPS-induced inflammation can alter the excitability of vagal nodose ganglion neurons [50,51]. Microbiota-derived metabolites are discussed in Section 3.

Short-chain fatty acids—mainly acetate, propionate, and butyrate—are produced by anaerobic fermentation of dietary fiber and resistant starch, and their levels can be modulated by altering the gut microbiota through probiotic intake [52,56,57,58,59]. Butyrate serves as a key energy source for colonic epithelial cells, upregulating tight-junction proteins (e.g., claudin, occludin) and maintaining barrier integrity [60], thereby limiting bacterial translocation into the circulation.

A subset of SCFAs crosses the BBB via monocarboxylate transporters [61,62], or acts on free fatty acid receptor 3 on brain endothelial cells [63,64,65], modulating BBB stress responses and permeability. Centrally, SCFAs regulate microglial maturation, homeostasis, and neuroinflammation [66]; butyrate additionally inhibits histone deacetylases [67,68,69] and is associated with increased brain-derived neurotrophic factor expression, synaptogenesis, and neurogenesis in preclinical models [52]. Overall, SCFAs link diet, microbiota, barrier function, and neural plasticity, representing intervention targets that require further validation in well-controlled human studies [70,71,72].

The gut microbiota both synthesizes neuroactive molecules and modulates their host production through substrate supply and metabolic regulation [39]. Figure 2 outlines key microbial sources, products, BBB permeability, and signaling routes.

Serotonin (5-HT). About 90–95% of total 5-HT is synthesized by intestinal enterochromaffin (EC) cells, and peripheral 5-HT does not cross the BBB. Spore-forming microbes promote gut-derived 5-HT synthesis and release by regulating EC cells, influencing motility, immunity, and behavior through peripheral and neural pathways [73,74,75,76,77].

γ-Aminobutyric acid (GABA). Lactobacillus, Bifidobacterium, and some Bacteroides produce GABA, whose central effects are likely mediated indirectly—via the vagus nerve, immune signaling, or barrier modulation—rather than by direct BBB passage [78].

Glutamate and dopamine-related signaling. Certain strains convert glutamate to GABA, shaping the peripheral substrate balance between excitation and inhibition. Peripheral dopamine poorly crosses the BBB, but its precursors—levodopa (L-DOPA) and large neutral amino acids (tyrosine, phenylalanine)—enter via large neutral amino acid transporter 1 (LAT1), influencing central dopaminergic activity [79,80].

From an interventional standpoint, targeting measurable metabolites or pathways may be more practical than manipulating the microbiota as a whole—for example, enhancing SCFA production through fiber intake or modulating 5-HT precursor availability via strain selection and nutrition. These approaches remain under study, and no definitive human evidence yet supports their efficacy for core ASD symptoms [16].

Tryptophan is metabolized through three main routes: the serotonin, kynurenine (KYN), and indole pathways [47]. Inflammation-induced activation of indoleamine 2,3-dioxygenase (IDO) and tryptophan 2,3-dioxygenase (TDO) shifts metabolism toward the KYN pathway, generating metabolites such as quinolinic acid and KYN, which exert N-methyl-D-aspartate receptor-related neurotoxic and immunomodulatory effects [47,81].

In ASD, peripheral tryptophan (TRP), KYN levels, and their ratios (e.g., KYN/TRP) show inconsistent findings. Although some studies report KYN-pathway activation alongside low-grade inflammation, recent systematic reviews and meta-analyses have not confirmed consistent alterations in peripheral TRP metabolism or tryptophan catabolites. Thus, KYN-pathway imbalance remains a plausible but non-universal mechanism, potentially amplifying gut–immune–brain signaling in subsets of individuals [81,82,83].

Many studies report gut microbiota differences between individuals with ASD and neurotypical controls [2], yet cross-study consistency remains limited. Some cohorts note phylum- or genus-level shifts—such as altered Firmicutes-to-Bacteroidetes ratios [84] and increased Proteobacteria or Sutterella—while others fail to replicate these patterns [2,85]. No single, universal "ASD microbiome signature" has been identified [86], and results depend heavily on sample source and methodology [87,88,89]. Table 2 summarizes taxa-level findings in ASD, highlighting recurrent trends, inconsistent results, and key confounders across studies.

Drivers of divergence include:(1)Clinical and behavioral heterogeneity: ASD phenotypes and the spectrum of GI comorbidities vary widely, with pronounced subgroup differences [53].(2)Lifestyle and confounding factors: notably restrictive diets and food selectivity, which can shape microbial differences and correlate with ASD features [86,89].(3)Sequencing and analytic variation: 16S rRNA gene amplicon sequencing versus whole-metagenome shotgun sequencing differ in taxonomic resolution and functional inference; batch effects, often addressed with tools like ComBat-Seq or RUV-III-NB, and statistical pipelines, which must account for the compositional nature of microbiome data using methods such as centered/additive log-ratio transformations, ANCOM-BC, or ALDEx2, further influence differential-abundance conclusions [89,106,107,108].

Against this backdrop, research focus is shifting from identifying specific taxa toward mapping functional pathways and metabolic networks. Multi-omics integration suggests that reproducible ASD-associated signals often involve amino acid, carbohydrate, and lipid metabolism, correlating with host transcriptomic and dietary profiles. These findings are largely correlational and remain mechanistic rather than causal [108].

Large-scale, multi-domain metagenomic studies employing machine learning have identified discriminative microbial and functional features—spanning bacteria, archaea, fungi, viruses, and pathways—with within-cohort classification AUCs up to ~0.9. However, external validation, cross-cohort reproducibility, and clinical translation remain major challenges [109].

Dysbiosis and impaired epithelial or BBB integrity are candidate mechanisms linking gut and brain in ASD. Loss of mucosal homeostasis allows translocation of bacterial components (e.g., LPS) and metabolites (e.g., p-cresol) into the circulation, eliciting systemic inflammation that can influence BBB permeability and central immune activity [110]. Elevated urinary p-cresol has been repeatedly reported, though its causality and diagnostic value remain uncertain [111].

A Neuroimaging and biofluid studies further indicate neuroimmune activation: increased 18-kDa translocator protein binding and higher inflammatory mediators in cerebrospinal fluid have been observed, albeit inconsistently and with limited specificity [112,113,114].

Collectively, these findings support a plausible cascade—gut inflammation → barrier dysfunction → systemic cytokines → BBB alteration → glial activation—potentially linking gastrointestinal and behavioral phenotypes in subsets of individuals with ASD. Prospective, controlled studies integrating dietary, pharmacologic, and omic variables are required to determine causality and modifiability [108].

Current evidence consistently demonstrates associations among ASD, gut microbiota composition, gastrointestinal symptoms, and behavioral phenotypes. Yet, most studies are cross-sectional or retrospective and therefore cannot infer causality [75]. A central confounder is selective and hypersensitive eating: many children with ASD exhibit sensory aversions to taste, texture, and smell, leading to restricted diets and nutrient imbalances [53]. Such dietary patterns alone can reshape gut microbial and metabolic profiles [86]. Hence, observed microbial alterations may reflect secondary effects rather than etiologic drivers. Robust causal inference will require designs that rigorously control for diet, medication use, and comorbidities [89,115].

MR uses genetic variants robustly associated with an exposure (e.g., microbial taxa or functions) as instrumental variables to infer causal effects under specific assumptions, thereby mitigating confounding and reverse causation [116,117]. Recent two-sample and bidirectional MR analyses have reported directional links between selected taxa (e.g., Turicibacter, Prevotellaceae) and ASD, although directions and effect sizes remain inconsistent across studies; some even suggest potentially protective associations [117]. These findings are constrained by weak instruments in microbiome GWAS, horizontal pleiotropy, and sample overlap, necessitating sensitivity analyses (e.g., weighted median, MR-Egger regression, MR-PRESSO, Steiger tests) and colocalization to reduce bias. Overall, MR currently offers preliminary genetic evidence rather than definitive causal proof [117,118].

Multi-omics integration—including metagenomics, transcriptomics, metabolomics, and host phenotypic data—shifts focus from individual taxa to functional pathways [108]. Large-scale integrative studies highlight alterations in amino acid, carbohydrate, and lipid metabolism, correlating with host brain-region transcriptomic profiles and dietary diversity [108]. However, these associations remain primarily correlational and require prospective and interventional validation.

Emerging analytical frameworks such as two-step mediation MR can decompose putative causal chains (e.g., "microbiota → circulating metabolite → ASD"), while reverse MR ("ASD → microbiota") helps assess bidirectionality and mediation, informing mechanistic targeting of metabolic and dietary pathways [119].

A credible research trajectory would therefore include: (i) prospective cohorts with harmonized data on diet, medication, and behavior analyzed using longitudinal and cross-lagged models; (ii) multivariable MR with colocalization and multi-omics integration to prioritize causal pathways; and (iii) interventional validation through randomized controlled trials (RCTs) enhancing dietary diversity or targeting metabolites and pathways—explicitly accounting for the bidirectional gut–brain–behavior feedback loop and integrating gut-targeted with behavioral-nutritional strategies [89,108]. This comprehensive approach requires a systematic framework that outlines not only the validation of forward causal pathways (microbiome → ASD), but also the investigation of reverse causation (ASD → microbiome), thereby fully addressing the system's bidirectional dynamics (see Box 1).

Given recurrent evidence of MGBA imbalance, probiotics and prebiotics have been evaluated as microbiome-targeted interventions in multiple clinical studies [120,121,122]. Systematic reviews and RCTs indicate that these interventions can alleviate GI symptoms in some children and, for certain strains or consortia, improve selected behavioral measures [120,123,124,125]. Other reviews, however, report inconsistent or nonsignificant effects on core ASD symptoms such as social communication and restricted, repetitive behaviors [105]. Overall, the evidence remains heterogeneous, with small and strain-specific effects influenced by dose, duration, baseline GI symptoms, and diet [126,127].

Methodologically, most studies are limited by small samples, heterogeneous interventions, nonstandardized outcome measures, and short follow-up. Open-label designs are prone to placebo and observer bias, and even RCTs show reduced comparability due to differences in strains, consortia, and baseline GI status [105]. A cautious summary is that evidence supporting improvements in GI symptoms is relatively consistent, whereas effects on core behavioral symptoms remain suggestive [126].

Recognizing ASD heterogeneity [86], the concept of personalized or precision synbiotics—tailoring strain–prebiotic combinations to individual microbiome and symptom profiles—is emerging. Pilot and open-label studies show modulation of microbial function and GI outcomes, with limited behavioral improvement, but large-scale, double-blind multicenter trials with standardized endpoints are lacking. Thus, personalized synbiotics currently represent a translational research direction rather than a clinical therapy [128].

As a more potent modality for gut ecosystem remodeling, FMT has shown encouraging but preliminary evidence in ASD. Open-label microbiota transfer therapy studies, such as those by Kang et al., reported improvements in pediatric GI symptoms (≈59%) and reductions in Childhood Autism Rating Scale scores (≈47%) sustained over two years [137]. However, these results derive from small, uncontrolled studies and require confirmation in double-blind RCTs [138,139].

Recently, randomized, double-blind, placebo-controlled, and multicenter trials of oral FMT have emerged, elevating the level of evidence. Overall, findings suggest potential benefits for GI outcomes and certain behavioral measures, though effects vary by scale, time point, and cohort. Larger samples and standardized endpoints are needed to clarify reproducibility, effect size, and durability [27,140,141].

Safety and regulatory oversight warrant particular attention. The U.S. Food and Drug Administration (FDA) has issued safety alerts—most notably in 2019 and 2020— regarding the transmission of multidrug-resistant organisms, opportunistic pathogens (e.g., Escherichia coli, Enterococcus faecium), and rare cases of severe infections and deaths resulting from donor stool–derived FMT materials [142]. These events underscored the need for systematic donor screening, including comprehensive medical history, stool and blood testing for infectious agents, and specific screening for antimicrobial resistance genes or MDRO carriage. Outside of recurrent Clostridioides difficile infection (rCDI)—the only indication currently approved for clinical FMT—such interventions are considered investigational and must adhere to regulatory pathways such as FDA Investigational New Drug applications or equivalent local ethical oversight. At present, approved microbiome-based products—Rebyota (enema formulation) and Vowst (oral spore-based therapy)—are indicated solely for rCDI prevention and are not generalizable to ASD [143]. Before clinical adoption, FMT for ASD should be validated through large, multicenter, double-blind RCTs using standardized donor screening, preparation, delivery, and follow-up protocols, to robustly establish both efficacy and safety [139].

The MGBA represents a conceptual framework for exploring part of the clinical heterogeneity observed in ASD. Gastrointestinal symptoms appear more prevalent among individuals with ASD and are frequently accompanied by shifts in microbiome composition and metabolite profiles, which are broadly consistent with mechanistic hypotheses involving neuroimmune and neuroplasticity pathways. Together, these findings form a convergent yet largely associational body of evidence [144]. However, no reproducible or universal "ASD microbiome signature" has been established across cohorts; reported taxa-level alterations often vary in both direction and magnitude among studies, reflecting substantial methodological and population heterogeneity. The most reproducible signals tend to converge on microbial functional pathways rather than on specific taxa, suggesting that pathway-level features may better capture biologically relevant host–microbe interactions. Causal directionality has not yet been established, and major confounders—including selective eating, medication use, and methodological variability—persist [21,75,108]. A recent large-scale, multi-domain metagenomic study identified functional signatures with potential discriminatory capacity, indicating a step toward translational exploration. Nonetheless, these models remain at an early research stage and require rigorous prospective validation, standardized data collection, and cross-population reproducibility assessment before any clinical application [109].

The primary contribution of this review is translating abstract concepts of causal inference and precision-oriented study design into actionable frameworks for the ASD-MGBA field. By synthesizing complex methodological challenges and proposing concrete, actionable frameworks (e.g., the bidirectional causality model inand the confounder control strategies in), this work provides a specific roadmap for future research. Box 1 Box 2

However, this review has limitations. As a narrative review, the selection of studies is not exhaustive and is subject to the authors' selection bias, unlike a systematic review or meta-analysis, which quantifies all available evidence. The forward-looking proposals, particularly for biomarker cutoffs and patient stratification, are intended as conceptual examples to guide future trial design rather than as definitive clinical guidelines.

Future investigations should aim to move from correlation toward mechanistic understanding and causal inference. In addition to advancing from association to causation, future research must rigorously address key methodological confounders that may obscure microbiome–behavior relationships (). Box 2

Studies need to link specific microbial metabolites and pathways to neural plasticity through controlled animal and organoid models, complemented by longitudinal, standardized multi-omic human cohorts that rigorously account for diet, medication, and geographic variation. Methodologically, these multi-center studies should model as a random effect and employ leave-site-out validation for robustness, while multi-omics integration can be achieved using frameworks such as MOFA+ or DIABLO. MR analyses should be employed cautiously—only when instrumental variable assumptions are satisfied—and their limitations transparently reported. Therapeutic trials should move beyond a one-size-fits-all approach by integrating gut-targeted, behavioral, and nutritional interventions. A critical next step is to stratify participants based on well-defined clinical and biological phenotypes to identify subgroups most likely to benefit from MGBA-targeted therapies. For instance, future RCTs could prioritize recruitment of specific subgroups, such as individuals with (1) a high burden of GI symptoms, (2) pronounced restrictive eating behaviors or sensory hypersensitivities that directly shape the microbiome, or (3) a recent history of significant microbial disruption, such as exposure to antibiotics or PPIs.

To objectify this stratification, a candidate biomarker panel could be employed for patient screening and monitoring. Such a panel might include (1) microbial metabolites like SCFAs and the tryptophan–kynurenine ratio to assess metabolic function; (2) gut-derived xenobiotics like p-cresol, which reflect dysbiotic activity; (3) markers of gut inflammation, such as fecal calprotectin; and (4) key systemic inflammatory cytokines. This approach would enable the development of more precise protocols. For example, a trial might set an inclusion criterion of fecal calprotectin > 50 µg/g to target gut inflammation, or it could involve longitudinal monitoring of the tryptophan-kynurenine ratio at 3- or 6-month intervals to gauge therapeutic response to an immunomodulatory probiotic. These trials must be conducted as preregistered, multicenter RCTs with harmonized endpoints to ensure reproducibility and clinical relevance.

Particular caution and regulatory oversight are essential for FMT, which remains experimental outside of Clostridioides difficile infection. Ultimately, reproducibility, transparent data sharing, and formal causal inference frameworks will determine whether functional microbiome and metabolite patterns can be validated as reliable biomarkers or as modifiable contributors to ASD-related phenotypes.