Identifying gut microbiota composition disparities in autistic individuals and their unaffected siblings: correlations with clinical characteristics

Abstract

Recent research has begun to illuminate a potential link between autism spectrum disorder (ASD) and the microbial environment. The unaffected sibling design is a valuable approach to identifying markers and protective factors for the condition, particularly in Asian populations lacking specific data. We assessed 239 autistic individuals, 102 unaffected siblings (SIB), and 81 typically developing controls (TDC) aged 4 to 25 years. We analyzed fresh stool samples via 16S rRNA amplicon library preparation and Illumina V3V4 sequencing. We employed taxonomic diversity analysis and microbiota differential abundance analysis to discern variations in microbial composition. In addition, we analyzed the associations between microbiota profiles and autistic symptoms, emotional/behavioral problems, and gastrointestinal symptoms. The SIB had higher alpha diversity, and autistic individuals had a different beta diversity compared to the TDC. Compared to ASD and SIB, the TDC group exhibited a higher relative abundance of microbiota, including Blautia, Eubacterium hallii group, Anaerostipes, Erysipelotrichaceae UCG 003, Parasutterella, and Ruminococcaceae UCG 013 at the genus level. Furthermore, the family Prevotellaceae and genera Agathobacter were predominant in SIB compared to ASD and TDC. We found that the microbial communities were related to autistic symptoms and gastrointestinal symptoms. Moreover, individuals with more Anaerostipes exhibited significantly less social impairment and internalizing problems. Our study reveals unique microbial compositions in the ASD and SIB groups and a relationship between behavior patterns and microbial composition. These findings suggest the potential of microbial interventions for autistic individuals that warrant further exploration.

Introduction

Autism spectrum disorders (ASD) are complex neurodevelopmental conditions characterized by social interaction and communication impairments and restricted/repetitive interests/behaviors [1]. The prevalence rate of ASD has risen dramatically on a global scale, with a substantial increase observed from 1 in 150 children in the year 2000 to 1 in 36 children by 2020 in the United States [2]. The prevalence has been markedly increasing in Asia, particularly in the East Asian region, where it has reached 0.51% [3] and 1% in Taiwan [4]. Autistic children commonly experience gastrointestinal issues (23% to 70%), like constipation, abdominal pain, gastroesophageal reflux, flatulence, diarrhea, vomiting, and nutrition problems [5–8], which may increase tantrums, aggression, self-injury, and sleep disturbances [6, 7]. Research shows a direct correlation between the severity of autistic symptoms and gastrointestinal symptoms [9, 10], suggesting a potential gastrointestinal contribution to ASD [11, 12]. Emerging studies focus on a possible connection between ASD and the gut microbiota [13], exploring how altering gut microbiota might ameliorate autistic symptoms [14, 15]. Kang et al. conducted fecal microbiota transplantation in children with ASD, altering the gut ecosystem and improving gastrointestinal and autistic symptoms [16]. Their 2-year follow-up study confirmed sustained benefits for both autistic symptoms and gut microbiota stability [17]. This connection implies that gut microbiota could influence brain development with a bidirectional interaction [18, 19], highlighting its role in ASD pathology and potential treatment strategies for both gastrointestinal and autistic symptoms [20].

While previous research has linked gut microbiota to the etiology of ASD, identifying specific bacteria remains inconsistent. Selecting appropriate control groups is crucial, as various lifestyle factors can significantly influence gut microbiota composition [21, 22]. Using an unaffected sibling design, our study minimized the impact of diet and lifestyle differences to uncover potential markers and protective factors associated with the disorder. This sibling-based approach ensured higher similarity in diet, environment, lifestyle, and genetics, crucial for understanding microbiota influences in ASD [23, 24].

Studies on microbiota changes in ASD using unaffected sibling (SIB) controls showed inconsistent results. Some research indicates increased species richness in relatives of ASD compared to controls [25], while others report no significant differences [26–29]. An Italian study noted specific microbial composition changes in siblings relative to those with ASD. In contrast, other studies found increased Lactobacillus, Desulfovibrio, and Clostridia in autistic children but decreased Bifidobacterium compared to siblings and typically developing controls (TDC). As for the microbial composition, an Italian study (10 ASD, 10 SIB) found specific microbial composition changes (increases or decreases) in SIB relative to ASD [25]. In contrast, in a small study (n = 29 in total), Tomova, Husarova, Lakatosova, Bakos, Vlkova, Babinska [14] found increased Lactobacillus, Desulfovibrio, and Clostridia species but decreased Bifidobacterium species in ASD compared to SIB and TDC. Some studies suggest Bifidobacterium might have a protective role in ASD due to its elevated level in SIB compared to autistic children [30]. However, several studies found no significant gut microbiota differences between autistic children and their siblings [28, 29]. These variations in findings could be explained by small sample sizes [14, 25], sampling methods [31], diverse phenotyping, gut microbiota profiling techniques [32], and geographical differences [33].

Research on ASD comparing autistic individuals, their SIB, and TDC are scarce and inconclusive, characterized by small sample sizes. This is particularly true for East Asian populations, where there is a significant lack of detailed clinical data despite the significant influence of geographical and cultural factors in shaping microorganism composition in such large populations. In addition, the role of gut bacteria in the disorder’s pathophysiology remains limited [21]. Our study aims to fill these gaps by investigating associations between the microbiota and clinical features and identifying the microbial profiles among autistic individuals, their SIB, and TDC in East Asia. We seek to understand how gut microbiota patterns might relate to autistic symptoms. We hypothesize variations in gut taxonomic diversity across these three groups (i.e., ASD, SIB, and TDC). Additionally, we anticipate observing unique microbial taxonomic compositions in ASD and SIB compared to TDC.

Methods

Participants and procedure

A total of 422 individuals, aged 4 to 25 years, were enrolled in the study: 239 autistic individuals (mean age: 12.13 ± 6.19 years) from the National Taiwan University Hospital, Taipei, Taiwan,102 biological SIB of autistic individuals (mean age: 12.01 ± 5.78 years), and 81 TDC participants (mean age: 13.77 ± 5.92 years). Autistic individuals were clinically diagnosed with ASD according to the DSM-5 criteria [34] and confirmed through the Autism Diagnostic Interview-Revised (ADI-R) [35] and the Autism Diagnostic Observation Schedule (ADOS) [36] interviews, utilizing authorized Chinese editions of ADI-R and ADOS by the World Psychiatric Association in June 2007 and April 2008, respectively. Both are to assist ASD diagnosis in ethnic Chinese populations, as previously documented [36–38]. Additionally, we converted the ADOS-2 algorithm total raw score proposed by Gotham et al. [39] into a standardized calibrated severity score [40, 41]. The SIB was assessed by the Mandarin version of the Kiddie-Schedule for Affective Disorders and Schizophrenia for School-Age Children–Epidemiological Version interview (K-SADS-E) [42, 43] and confirmed as non-autistic through clinical diagnostic interviews conducted by the corresponding author, S.S.-F.G.

TDC participants, recruited from the same neighborhoods as autistic participants, had no lifetime or current DSM-5 psychiatric disorders, as made by the K-SADS-E. The exclusion criteria applied to all participants included neurological or systemic medical conditions, significant mental disorders (such as substance use disorders, schizophrenia, mood disorders, and anxiety disorders), current use of psychotropic medication, recent probiotics or prebiotics consumption, or antibiotic treatment within the past month, and a full-scale intelligence quotient (FIQ) < 70, as assessed by the Wechsler Intelligence Scale for Children–3^rd Edition (age < 16 years) or Wechsler Adult Intelligence Scale-3^rd Edition (age ≥ 16 years). Given the potential influence of gastrointestinal symptoms and dietary preferences on gut microbiota diversity [21], we evaluated gastrointestinal issues using the Gastrointestinal Symptoms Questionnaire [44]. We ensured that all participants had not been diagnosed with gastrointestinal disorders in the preceding three months and had refrained from using probiotic supplements. For participants who maintained consistent medication usage and regular dietary and bowel habits, we considered their gastrointestinal habits and problems over the past four weeks. All participants were recruited between August 1, 2017, and December 12, 2022.

We obtained written informed consent from all participants and their parents after a detailed explanation of the study’s purpose and procedures. All experimental procedures and data collection methods were conducted in strict accordance with relevant institutional and international ethical standards and regulations, including the Declaration of Helsinki and the guidelines of the local institutional review board. The study was approved by the Research Ethics Committee of the National Taiwan University Hospital, Taipei, Taiwan (approval number 201707041RINA; ClinicalTrials.gov numbers NCT02719067 and NCT04873674).

Measures

Social responsiveness scale (SRS)

The SRS is a self-report or caregiver-report questionnaire frequently used to quantify the frequency of autism-related behaviors [45], unaffected by age, race, IQ, or education level [46]. The SRS is a 65-item scale using a 4-point scale from 0 (never true) to 3 (almost always true). A higher SRS score indicates more substantial characteristics associated with autism. The Chinese SRS exhibits good reliability and validity, with fair internal consistency (Cronbach’s α = 0.94–0.95), reliability (intra-class correlation, ICC = 0.75–0.85), and convergent validity as demonstrated by the Chinese Social Communication Questionnaire (Pearson’s r = 0.61–0.87) [47]. The Chinese SRS has been widely used in evaluating social deficits in Taiwan [48, 49].

Child behavior checklist (CBCL)

The CBCL is a widely used questionnaire that parents or caregivers complete to assess emotional and behavioral issues in children and adolescents. It consists of 118 items, each rated on a scale of 0 (untrue), 1 (somewhat or occasionally true), or 2 (very or frequently true). The CBCL includes eight domains: Anxious/Depressed Symptoms, Attention Problems, Aggressive Behaviors, Delinquent Behaviors, Social Problems, Thought Problems, Somatic Complaints, and Withdrawn [50] and two subscales: internalizing (emotional) problems and externalizing (behavioral) problems, as outlined by Pandolfi, Magyar and Dill [51]. Based on Taiwanese child [52] and adolescent [53] norms, raw scores on these subscales are converted to T-scores, with a mean of 50 and a standard deviation of 10. The combined T-scores for Anxious/Depressed, Aggression, and Attention subscales, or CBCL-DP [54], reflect self-regulation across affect, behavior, and cognition. CBCL-DP is a widely used measure of emotional and behavioral dysregulation in clinical and non-clinical populations [55]. Elevated CBCL-DP composite T-scores in children with ASD have been associated with increased clinical severity and subsequent psychosocial impairments in previous research [56–58]. The CBCL was originally designed for participants aged 6–18 years, and we applied it to all participants in the current study to ensure consistency in emotional and behavioral assessment across age groups.

Gastrointestinal symptoms questionnaire

The Gastrointestinal Symptoms Questionnaire, developed by Bovenschen, Janssen, van Oijen, Laheij, van Rossum and Jansen [44], assesses the severity of gastrointestinal symptoms experienced in the past four weeks. It includes 32 items, categorized into four subscales: abdominal pain (6 items), upper GI symptoms (11 items), bowel habits (11 items), and food picking (4 items). We interviewed participants and their parents to rate each item on a scale of 0 to 6, where 0 represents “no complaints,” and 6 indicates the most severe and intolerable symptom.

Collection of fecal samples and extraction of DNA

We collected fresh fecal specimens from all participants and immediately deposited them into sterile collection containers. These specimens were then temporarily preserved through freezing at −20 °C for a maximum period of seven days. Subsequently, the samples were transported to the laboratory within two hours using dry ice containers. Upon arrival at the laboratory, the specimens were stored at −80 °C for later analysis. The fecal sample processing consisted of extracting genomic DNA, PCR amplification and purification, and library preparation for subsequent sequencing.

Extraction of genome DNA

Genomic DNA was extracted from the samples using a column-based extraction technique (QIAamp PowerFecal DNA Kit, Qiagen). The concentration of the extracted DNA was quantified using a NanoDrop Microvolume and subsequently adjusted to 5 ng/μl for further processing.

PCR amplification and purification

The 16S rRNA gene sequencing targeted the V3-V4 region, which was amplified using specific primer sets (341 F: 5’-CCTACGGGNGGCWGCAG-3’, 806 R: 5’GACTACHVGGGTAT CTAATCC -3’), following the 16S Metagenomic Sequencing Library Preparation protocol (Illumina). Briefly, NanoDrop with 12.5 ng of genomic DNA was utilized for the polymerase chain reaction (PCR), carried out with KAPA HiFi HotStart ReadyMix (Roche) under the following conditions: 95 °C for 3 min; 25 cycles of 95 °C for 30 s, 55 °C for 30 s, 72 °C for 30 s; followed by 72 °C for 5 min and a hold at 4 °C. The PCR products were monitored on a 1.5% agarose gel, and samples exhibiting a distinct band around 500 base pairs were selected and purified using AMPure XP beads for subsequent library preparation.

Library preparation

The sequencing library preparation followed the 16S Metagenomic Sequencing Library Preparation protocol (Illumina). A secondary PCR was carried out utilizing the 16S rRNA V3-V4 region amplicon and the Nextera XT Index Kit, which incorporated dual indices and Illumina sequencing adapters (Illumina). The quality of the indexed PCR products was evaluated using the Qubit 4.0 Fluorometer (Thermo Scientific) and the Qsep100™ system. Equal quantities of the indexed PCR products were combined to create the sequencing library. Ultimately, the library was sequenced on the Illumina MiSeq platform, generating paired 300-base pair reads. The settings for the second PCR were based on the Illumina standard procedure [59].

Amplicon sequence variants sequencing

Amplicon sequencing was conducted on 300 bp paired-end raw reads, and each sample was demultiplexed based on barcode identification. Subsequently, primer and adapter sequences were removed from the paired-end reads using the QIIME2 plugin with standard settings [60]. Amplicon sequence variants (ASVs) were generated through a denoising pipeline using the QIIME2 DADA2 plugin (v2021.4; https://qiime2.org/)), encompassing quality filtering, dereplication, dataset-specific error model learning, denoising, joining of full-length paired-end reads, and chimeras removal [61, 62]. The full length of both forward and reverse reads was used to generate the finalized reads. Trimming and filtering were executed with a maximum of two expected errors per read (maxEE = 2). The DADA2 algorithm accurately merged paired-end reads with overlapping 12 base pairs at near-zero error rates. Taxonomic classification for each representative sequence was performed using the feature classifier with the RDP classifier algorithm (v. 2.2) [63, 64] and the algorithm within QIIME2 [65], leveraging information from the Silva database (v. 132). Multiple sequence alignment was conducted using the QIIME2 alignment MAFFT [66] against the Silva database [67–69] to assess sequence similarities among ASVs. A phylogenetic tree was constructed using the QIIME2 phylogeny fasttree [63, 64] with a set of representative ASV sequences. To address variations in sequence depth across samples, the abundance information of ASVs was normalized by rarefying to the minimum sequence depth using the QIIME2 script (qiime feature-table rarefy).

Statistical analysis

Alpha and beta diversity

This study assessed alpha diversity in microbial samples using the QIIME2 pipeline and metrics like Shannon and Simpson indices [65]. It evaluated community richness using Chao1 and ACE indices, and the relative abundance and evenness, which account for diversity, were assessed by Pielou’s evenness, Shannon, and Simpson indices. Beta diversity was analyzed using PERMANOVA (Permutational Multivariate Analysis of Variance) to assess overall differences in bacterial composition between different groups, followed by pairwise PERMANOVA. The QIIME2 pipeline was utilized to compute the weighted and unweighted UniFrac and Bray-Curtis dissimilarities [66, 67].

Ordination analysis

Principal Coordinate Analysis (PCoA) was used for intricate and multidimensional data visualization, using a distance matrix to derive principal coordinates [68]. The distance matrix, encompassing weighted and unweighted UniFrac and Bray-Curtis dissimilarities among all samples, was transformed to yield a novel set of orthogonal axes. The initial principal coordinate represented the most influential variable, followed by the second principal coordinate, and so forth. The PCoA analysis was executed in R, utilizing the stat and ggplot2 packages. In our statistical analysis, we used a zero-inflated Gaussian log-normal model for differential abundance analysis [69] within the Bioconductor metagenomeSeq package. Additionally, we performed Welch’s t-test with STAMP software (version 2.1.3) [70]. Finally, sample and group comparisons with categorical metadata were made using non-parametric multivariate methods, including analysis of similarities (Anosim) [71, 72], multiple response permutation procedure (MRPP) [73], and permutational multivariate analysis of variance (Adonis) [74, 75] based on Bray-Curtis dissimilarity. Each method provides complementary strengths: ANOSIM ranks dissimilarities to detect differences in community structure, MRPP evaluates the degree of group separation, and ADONIS quantifies the variance attributable to group differences.

Biomarker identification and taxonomic analysis

We used the linear discriminant analysis effect size (LEfSe) method to identify significant biomarkers by detecting bacterial taxa with notably different relative abundances between groups [76]. This involved using a non-parametric Wilcoxon rank-sum test followed by linear discriminant analysis (LDA) to evaluate the effect size of these taxa. Taxa with an LDA score (log 10) above 2 were deemed statistically significant. We applied the Benjamini & Hochberg procedure to adjust for false discovery rates (FDR) in multiple comparisons [77], considering results with FDR-adjusted p-values (pFDR) of ≤ 0.05 as statistically significant.

Functional prediction

Our functional analysis used 16S rRNA sequencing data and the Phylogenetic Investigation of Communities by Reconstruction of Unobserved States 2 (PICRUSt 2) to predict functional genes based on bacterial gene sequences [78]. The accuracy of these predictions was assessed using the Nearest Sequenced Taxon Index, with lower values indicating higher accuracy. Additionally, we employed BugBase to predict microbial community phenotypes across six distinct categories, encompassing Gram staining, oxygen tolerance, biofilm formation, mobile element content, pathogenicity, and oxidative stress tolerance [79].

Association between microbiota abundance and clinical manifestations

To delineate the associations between microbiota profiles and clinical manifestations, we used the Mantel test [80], based on a distance matrix, to investigate the relationships between clinical factors (autistic symptoms, emotional/behavior problems, and gastrointestinal (GI) symptoms) and the overall microbial community, involving calculating distance attributions and converting them into a distance matrix for analysis [81, 82]. We utilized the Bray-Curtis distance matrix for the microbial community identified by the LEfSe test and created an Euclidean distance matrix for clinical variables. Subsequently, these two matrices were compared using Spearman correlations with 9999 permutations. Additionally, we conducted partial correlation analyses to examine links between microbial taxonomy relative abundance and clinical features, adjusting for age, sex, FIQ, and Body Mass Index (BMI).

Results

The sample characteristics

The demographics, intelligence, and clinical features of the ASD, SIB, and TDC groups are summarized in Table 1. The three groups had comparable age distributions (mean age = 12.42 ± 6.06 years) and showed no significant differences in BMI (mean = 18.69 ± 4.22 kg/m²). The proportion of males was higher in ASD (83.68%) than in TDC (66.67%) and SIB (50.98%). Both the SIB and TDC groups demonstrated significantly higher FIQ than ASD, with no significant differences between SIB and TDC. The ASD group had significantly higher SRS total and subdomain scores (Table 1) and more emotional/behavioral symptoms (Supplementary Table 1) than the SIB and TDC groups. No significant group differences were observed in GI symptoms, including abdominal pain, epigastric pain, upper GI symptoms, and bowel habits. However, the ASD group displayed stronger food preferences, especially for vegetables, fruits, and grains, compared to their TDC and SIB counterparts (Supplementary Table 2).

Table 1.

Demographics, intelligence, and clinical features between the ASD, SIB and TDC groups.

	ASD (n = 239)	SIB (n = 102)	TDC (n = 81)	One-Way ANOVA (Welch’s)		Post hoc Comparisonb
				F	p values
Age (years)	12.13 (6.19)	12.01 (5.78)	13.77 (5.92)	2.62	0.075	–
	range: 4–25	range: 4–25	range: 4–25
Gender (%)					<0.001a
Male	200 (83.68)	52 (50.98)	54 (66.67)
Female	39 (16.32)	50 (49.02)	27 (33.33)
BMI (ASD = 208, SIB = 74, TDC = 73)	18.85 (4.34)	18.64 (4.35)	18.26 (3.70)	0.62	0.537	–
Full-scale Intelligence quotient	95.16 (20.76)	106.76 (14.46)	112.12 (13.35)	31.84	<0.001	ASD < SIB ASD < TDC
Social Responsiveness Scale (SRS)
Social communication	38.08 (15.82)	14.35 (17.18)	8.75 (8.02)	232.19	<0.001	ASD > SIB > TDC
Stereotyped behaviors	18.52 (7.50)	7.01 (8.56)	3.76 (4.34)	233.30	<0.001	ASD > SIB > TDC
Social awareness	21.12 (4.85)	12.91 (6.97)	11.20 (5.65)	129.01	<0.001	ASD > SIB ASD > TDC
Social emotion	12.33 (4.93)	5.20 (5.27)	3.79 (3.71)	151.20	<0.001	ASD > SIB ASD > TDC
Total score	90.05 (28.79)	39.46 (35.15)	27.52 (18.30)	263.73	<0.001	ASD > SIB > TDC
Current Autism Diagnostic Interview-Revised (ADI-R)
Language/Communication	13.20 (4.90)	–	–
Reciprocal social interactions	10.27 (4.66)	–	–
RRSB	3.18 (2.13)	–	–
Autism Diagnostic Observation
Social Affect	10.77 (4.18)	–	–
Restricted and Repetitive Behavior	2.19 (1.84)	–	–
Overall Total	12.96 (5.03)	–	–
Calibrated Severity Score	6.86 (2.21)	–	–

ASD autism spectrum disorder, SIB unaffected sibling, TDC typically developing control; RRSB restricted, repetitive, and stereotyped patterns of behavior; ANOVA analysis of variance; BMI body mass index.

^aChi-square test.

^bTukey’s honest significant difference test.

The taxonomic diversity between ASD, SIB, and TDC

The Venn diagram shows a total of 11158 ASVs, with unique observed 4162, 2238, and 975 ASVs in ASD, SIB, and TDC, respectively (Fig. 1A). They included 629 genus-level microbiota and 300 species-level microbiota. Significant group differences were observed in alpha diversity indexes: higher alpha diversity in SIB than TDC based on ACE, Chao 1, and Richness, and higher alpha diversity in SIB than ASD according to the Pielou and Shannon indexes (Table 2A). We performed linear regression analysis to control for potential confounding variables. These group differences remained significant after adjusting for age and upper GI symptoms through sensitivity analyses (Table 2B). The Firmicutes to Bacteroidetes ratio was higher in TDC than in SIB (p = 0.037, pFDR = 0.109; Fig. 1B). The top ten relative prevalences of the microbiota genera are Bacteroides, Bifidobacterium, Blautia, Escherichia Shigella, Faecalibacterium, Subdoligranulum, Eubacterium hallii group, Akkermansia, Alistipes, Collinsella, and other genera (Fig. 1C). A taxonomy tree constructed using GraPhlAn [83] visualizes this differentiation (Supplementary Figure 1). Furthermore, significant differences were noted in the cluster-level evolutionary heat map of the microbiota (Supplementary Figure 2). The three groups (pFDR <0.001) showed significant differences in beta diversity on both weighted and unweighted UniFrac distances. The ASD group exhibited the highest intra-group microbiota variation for both weighted and unweighted UniFrac distances, followed by the SIB and TDC groups. (Fig. 1D). Sensitivity analyses were performed by including age and upper GI symptoms as covariates (Supplementary Table 3). Similar group differences were observed, and age (Supplementary Figure 3) and upper GI symptoms (Supplementary Figure 4) were also added as confounding factors in the PCoA plot. We further investigated an analysis of similarities in the relative abundance of the microbiota community composition of beta diversity between ASD, SIB, and TDC. We assessed the statistical significance of microbial composition differences between these groups using the Anosim, MRPP, and Adonis methods (Supplementary Table 4). The results showed a higher level of between-group difference in ASD than TDC (A score = 0.0007, pFDR value = 0.044; F = 1.379, pFDR = 0.035), with no significant differences between SIB and TDC or between ASD and SIB.

Fig. 1. The overview of the microbiome profiles in ASD, SIB, and TDC.

A Venn Diagram of ASVs. The Venn diagram summarizes the total number of 4,162, 2,238, and 975 ASVs in ASD, SIB, and TDC, respectively. The blue oval represents the ASD group, and the pink and purple ovals separately represent the SIB and TDC groups. B The Firmicutes to the Bacteroidetes ratio. There was no group difference between ASD, SIB, and TDC using the analysis of variance (ANOVA) corrected by false discovery rate (FDR). C The top 10 taxa of relative abundance at the genus level. D Beta diversity of ASD, SIB, and TDC. Beta diversity is presented using box plots and principal coordinate analysis (PCoA) with both weighted and unweighted UniFrac distances. The ASD group had the highest intra-group microbiota variation for both weighted and unweighted UniFrac distances, followed by the SIB and TDC groups. The ellipses represent the 95% confidence interval. ASD autism spectrum disorder SIB unaffected sibling TDC typically developing control.

Table 2.

A; Alpha diversity indexes for the ASD, SIB, and TDC groups. B; Alpha diversity indexes for the ASD, SIB, and TDC groups: Linear regression analysis adjusted for age and upper gastrointestinal symptoms.

A
	ASD (n = 239)	SIB (n = 102)	TDC (n = 81)	p-value (ANOVA)	Post hoc Comparisona
ACE	201.20 (90.07)	224.94 (122.88)	189.93 (67.86)	0.034	SIB > TDC
Chao1	201.61(90.33)	224.58 (122.01)	190.19 (68.05)	0.039	SIB > TDC
Pielou	0.70 (0.08)	0.72 (0.07)	0.72 (0.06)	0.002	SIB > ASD, TDC > ASD
Richness	195.95 (86.32)	218.58 (114.66)	186.04 (65.30)	0.037	SIB > TDC
Shannon	5.24 (0.82)	5.49 (0.72)	5.42 (0.65)	0.012	SIB > ASD
Simpson	0.93 (0.07)	0.94 (0.04)	0.95 (0.04)	0.019	–

B
	Age (Estimate/p-value)	Upper GI symptoms (Estimate/p-value)	Group (Estimate/p-value)	Group Comparison
ACE	1.047/0.191	−0.355/0.724	−13.589/0.034	SIB > TDC
Chao1	1.048/0.19	−0.349/0.728	−13.510/0.034	SIB > TDC
Pielou	<−0.001/0.223	<−0.001/0.392	0.030/0.002	TDC > ASD
Richness	1.049/0.189	−0.354/0.724	−13.654/0.033	SIB > TDC
Shannon	<0.001/0.993	−0.004/0.432	0.122/0.012	SIB > ASD
Simpson	<−0.001/0.633	<−0.001/0.784	0.016/0.018	–

^aTukey’s honest significant difference test.

ASD autism spectrum disorder SIB unaffected sibling TDC typically developing control; GI: gastrointestinal.

The differential microbial abundance analysis between ASD, SIB, and TDC by LefSe

We comprehensively examined disparities within microbiota communities from the phylum to the species level across ASD, SIB, and TDC groups to uncover evolutionary markers associated with ASD by studying the gut microbiota. The LefSe analysis (LDA > 2) identified 24 potential biomarkers differentiating the groups (Supplementary Figure 5). The ASD group exhibited a higher abundance of the bacterial genus Coprobacter. Notably, the SIB group increased microbiota from the family Prevotellaceae to the genera Prevotella 7 and Alloprevotella, as well as Lachnospiraceae bacterium A2, Coprococcus 1, Lachnospiraceae NK4A136 group, and Agathobacter at the genus level, compared to ASD and TDC. Additionally, the significantly increasing microbiota from the phylum Euryarchaeota, order Deferribacterales, family Deferribacteraceae, and genus Mucispirillum, as well as orders Corynebacteriales and Methanobacteriales, family Methanobacteriaceae, and genus Ruminiclostridium 5 were more prevalent in SIB than in ASD and TDC. The TDC group exhibited significant increases in the relative abundance of Blautia, Eubacterium hallii group, Anaerostipes, Ruminococcaceae UCG 013, Parasutterella, Erysipelotrichaceae UCG 003, Eubacterium ventriosum group at the genus level, and Bacteroides caccae at the species level.

As indicated by LDA > 3, eight microbiota exhibited the most significant disparities in taxonomic composition among the ASD, SIB, and TDC groups, as depicted in the cladogram in Fig. 2A. Specifically, there was an increase in the family Prevotellacea and the genus Agathobacter in the SIB group compared to ASD and TDC. The TDC group displayed a higher relative abundance of Anaerostipes, Blautia, the Eubacterium hallii group, Ruminococcaceae UCG 013, Parasutterella, and Erysipelotrichaceae UCG 003 at the genus level than ASD and SIB. However, no potential biomarkers with LDA > 4 were identified.

Fig. 2. The differential relative abundance of gut microbiota compositions in the ASD, SIB, and TDC groups.

A Linear discriminant analysis (LDA) effect size test. The linear discriminant analysis effect size (LEfSe) analysis showed that the LDA score > 3 reaches significance. The left panel illustrates the evolution of composition between groups, and the right represents the significance level. The alphabet next to the name of microbiota corresponds to the evolutionary cladogram. B Effect plot for differential abundance analysis between ASD and TDC. The ANOVA-Like Differential Expression 2 (ALDEx2) analysis identified higher Erysipelotrichaceae UCG 003 at the genus level in TDC than in ASD. The expected value of the distributional difference between groups (Y-axis: median log2 btw-condition diff = median log₂ between-group difference) and the expected value of the overall variance (X-axis: median log₂ win-condition diff = median log₂ within-group difference) was calculated. The effect size was used to assess the reproducibility of differences between groups. C Volcano plot for the analysis of the composition of microbiomes (ANCOM) test. (1) The ANCOM analysis identified higher Prevotellaceae at the family level in SIB than in TDC. (2) SIB has higher Parasutterella abundance at the genus level than TDC. TDC had a higher relative abundance of Anaerostipes, Ruminococcaceae UCG 013, and Erysipelotrichaceae UCG 003 than ASD. (3) At the species level, the abundance of Bacteroides caccae was higher in ASD than in TDC. The X-axis, compositional CLR mean difference, is transformed from the species abundance of samples and plotted as the F-statistic score. The y-axis represents the W value. D Metagenome sequence analysis between ASD, SIB, and TDC groups. Metagenome sequence analysis directly compared the species-level microbiota after controlling for covariates and corrected the p-value using the false discovery rate (FDR). (1) At the genus level, Alloprevotella, Prevotella 7, Lanchnospiraceae UCG 006, and Faecalibaculum were significantly higher in SIB than in TDC. (2) At the species level, Bacteroides caccae was significantly higher in TDC than in ASD. ASD autism spectrum disorder SIB unaffected sibling TDC typically developing control.

Group-level differences identified by ANCOM, ALDEx2, and metagenome analysis

To explore group-level disparities across various microbiota levels, as indicated in principal component and LefSe analysis, we employed methods such as Analysis of Composition of Microbiomes (ANCOM), ANOVA-Like Differential Expression 2 (ALDEx2), and metagenome sequence analysis to verify our findings further. ANCOM, through centered log-ratio (CLR) and W statistics, identified group differences, while ALDEx2 estimated the effect size of these between-group differences for cross-validation. Taxa were considered significantly differentially abundant if they met the following criteria: ANCOM (W statistic > 0.6), ALDEx2 (pFDR < 0.1), and metagenome sequence analysis (pFDR < 0.05). In ALDEx2 analysis, we found Erysipelotrichaceae UCG 003 at the genus level exhibited marginal statistical differences between ASD and TDC (effect size = 0.300, pFDR = 0.058) (Fig. 2B). Notably, Prevotellaceae (family level) was more abundant in SIB than TDC (W > 0.7, mean difference (MD) = 1.142) (Fig. 2C1). At the genus level, SIB showed higher Parasutterella abundance than TDC (W > 0.6, MD = 1.080) (Fig. 2C2). TDC had an increased relative abundance of Anaerostipes (W > 0.6, MD = 0.811), Ruminococcaceae UCG 013 (W > 0.6, MD = 0.901), Erysipelotrichaceae UCG 003 (W > 0.9, MD = 1.610) than ASD (Fig. 2C3). Species-level analysis revealed greater Bacteroides caccae in TDC than in ASD (W > 0.7, MD = 0.960) (Figure 2C4). Metagenome sequence analysis revealed that SIB had a higher relative abundance of Alloprevotella (pFDR = 0.013), Prevotella 7 (pFDR = 0.038), Lanchnospiraceae UCG 006 (pFDR = 0.013), and Faecalibaculum (pFDR = 0.038) at the genus level than TDC (Figure 2D1). Both ANCOM and metagenome sequence analysis showed that TDC had more Bacteroides caccae at the species level than ASD (pFDR < 0.001) (Figure 2C4 & D2). We observed a high degree of concordance in the major findings, particularly regarding the enrichment of taxa from the Lachnospiraceae and Prevotellaceae families in SIB and TDC, respectively, compared to ASD. While not definitive proof of robustness, this consistency across multiple analyses provides a more comprehensive understanding of the observed differences.

The predicted function and pathway between ASD, SIB, and TDC

The bacterial diversity of the functional microbiota, including aerobic (pFDR = 0.669), anaerobic (pFDR = 0.862), Gram-negative (pFDR = 0.191), and Gram-positive (pFDR = 0.191), did not demonstrate group differences (Supplementary Figure 6). Additionally, we used PICRUSt2 to explore the predicted functions of the microbiota. First, we explored prokaryotic taxa differences and linked them to metabolic functions in ASD, SIB, and TDC. The overall functional structure via PICRUSt2 analysis and Wilcoxon test, based on the MetaCyc database, identified six pathways with uncorrected p-values below 0.01. No statistically significant differences were found between the ASD and SIB groups in functional pathways. Notably, the functional prediction for L-glutamate degradation V showed higher SIB levels than TDC (Supplementary Figure 7A). Five pathways, including the super pathway of heme biosynthesis from glycine, coenzyme M biosynthesis I, L-glutamate degradation V, L-rhamnose degradation II, and GDP-D-glycero-α-D-manno-heptose biosynthesis, exhibited a higher mean proportion in ASD than TDC (Supplementary Figure 7B). However, none of them remained significant after the FDR adjustment.

Microbial community correlations with autistic, emotional/behavioral, and GI symptoms

The Mantel test revealed statistically significant correlations between microbial communities, as identified by LEfSe, and various autistic symptoms, controlled for age, sex, FIQ, and BMI, including social communication (rho = 0.023, p-value = 0.032) and unique mannerism (rho = 0.030, p-value = 0.019, Table 3). The Mantel test also found correlations between microbial communities and total GI symptoms (rho = 0.057, p-value = 0.008), as well as specific domains, including abdominal pain (rho = 0.058, p-value = 0.017), epigastric pain (rho = 0.054, p-value = 0.038), upper GI symptoms (rho = 0.042, p-value = 0.038), and bowel habits (rho = 0.061, p-value = 0.007). However, no significant correlations were observed between the variables and emotional/behavioral problems (Table 3).

Table 3.

The Mantel test of correlations between microbial communities identified by LEfSe test (LDA > 3) and autistic symptoms and emotional/behavioral problems, and GI symptoms.

Environmental variables	Correlation rho	p-value
Social Responsiveness Scale
Social communication	0.023	0.032
Unique mannerism	0.030	0.019
Social awareness	−0.013	0.761
Social emotion	0.023	0.065
SRS total score	0.023	0.059
Child Behavior Checklist
Dysregulation profile	0.185	0.185
Internalizing	0.186	0.186
Externalizing	0.140	0.140
Gastrointestinal Assessments (GI)
Abdominal pain	0.058	0.017
Epigastric pain	0.054	0.038
Upper GI symptoms	0.042	0.038
Bowel habits	0.061	0.007
Food preferences	0.004	0.400
Total score	0.057	0.008

The correlation of microbial genera abundance and behavioral symptoms was controlled by age, sex, BMI, and FIQ. Bolded rho and p-values indicate statistical significance (p-value < 0.05). To get rho, we examined Spearman correlation via Mantel test between the microbial community structure and behavioral symptoms. We calculated the Bray-Curtis dissimilarity matrix for species abundance and Euclidean distance matrix for behavioral symptoms as x matrix and y matrix of the equation to enter in the Mantel test.

ASD autism spectrum disorder SIB unaffected sibling; TDC typically developing control LEfSe Linear discriminant analysis effect size LDA Linear discriminant analysis.

The taxonomy-relative abundance correlations with autistic, emotional/behavioral, and GI symptoms

We conducted a partial Spearman correlation analysis between the relative abundance of species in the Lanchnospiraceae family (i.e., Blautia, Eubacterium hallii group, Anaerostipes, Agathobacter at genus level) and family Prevotellaceae, identified by LDA 3 and the SRS total scores and internalizing, externalizing problem, and dysregulation profile in CBCL, and total GI symptoms, controlled for age, sex, FIQ, and BMI. We found a negative correlation between the relative abundance of Anaerostipes and SRS total scores (rho = −0.11, p-value = 0.022) (Fig. 3A) and its subscores including social communication (rho = −0.11, p-value = 0.026), unique mannerism (rho = −0.10, p-value = 0.041), social awareness (rho = −0.10, p-value = 0.048), and social emotion (rho = −0.11, p-value = 0.031). Similar negative correlations were observed with internalizing problems (rho = −0.13, p-value = 0.007) (Fig. 3B), dysregulation profile (rho = −0.12, p-value = 0.012), and total GI symptoms (rho = −0.09, p-value = 0.053; Fig. 3C and Supplementary Table 5).

Fig. 3. The partial correlation between the identified relative abundance of gut microbiota and clinical features and gestational tract symptoms.

A The relative abundance of Anaerostipes was negatively associated with SRS total scores. B The relative abundance of Anaerostipes was negatively associated with the internal scores of CBCL. C The relative abundance of Anaerostipes was negatively associated with GIS total scores. The analysis was controlled by age, sex, FIQ, and BMI. SRS social responsiveness scale CBCL child behavior checklist GIS gestational tract symptoms.

Discussion

This first East Asian ASD sibling study, incorporating extensive phenotyping and accounting for geographical and ethnic characteristics, compared the gut microbiota across the ASD, SIB, and TDC groups. We found that the SIB group showed higher alpha diversity than ASD and TDC, with ASD displaying greater diversity than TDC. Various independent contrast methods revealed that the TDC group had a higher relative abundance of Ruminococcaceae UCG 013, Parasutterella, and Erysipelotrichaceae UCG 003 at the genus level than ASD. Furthermore, both SIB and TDC groups exhibited an increased relative abundance of the genus belonging to the Lachnospiraceae family. Specifically, TDC had higher Blautia, Eubacterium hallii group, and Anaerostipes, while SIB had elevated Agathobacter, Lachnospiraceae bacterium A2, Lachnospiraceae NK4A136, and Coprococcus 1. Additionally, the family Prevotellaceae and its genera Prevotella 7 and Alloprevotella were more abundant in the SIB group compared to ASD. Additionally, a higher relative abundance of Anaerostipes was associated with fewer autistic symptoms, dysregulation profiles, and internalizing problems, as well as controlling for age, sex, FIQ, and BMI.

Alpha diversity, a dysbiosis indicator, typically associates lower alpha diversity with reduced community richness and vulnerability to disturbances. Our findings showed no significant alpha diversity differences between ASD and TDC, except for the Pielou index, which indicated higher alpha diversity in TDC compared to ASD. This finding aligns with our previous reports [84] and other studies [28, 30, 85, 86]. Additionally, we observed higher alpha diversity in SIB than in TDC; however, there were no differences between ASD and SIB, mirroring mixed results from previous research. Some studies report similar alpha diversity levels in autistic individuals and their siblings [27, 28, 30, 85], while one study suggests higher diversity levels in ASD than in SIB [25]. These inconsistencies might stem from small sample sizes in earlier studies, potentially leading to Type 1 errors. For example, the survey by Angelis et al. included only ten participants in the ASD and SIB groups.

Our results indicated the highest intra-group microbiota variation in ASD, followed by SIB and TDC, in both weighted and unweighted UniFrac distances. This higher intra-group microbiota variation in ASD than TDC corresponds with our previous research [84] and several other studies [87, 88], possibly reflecting differences in food habits and higher GI symptoms in ASD [89, 90]. Earlier studies with relatively smaller sample sizes showed no significant beta diversity differences between autistic individuals and their siblings [26–28]. However, our findings align with a recent study that noted distinct beta diversity among three groups using weighted and unweighted UniFrac measures [91].

Our findings of decreased abundance in the genus Ruminococcaceae UCG 013 and Erysipelotrichaceae UCG 003 levels in TDC are consistent with prior findings [84]. We also observed a greater relative abundance of the genera within the Lachnospiraceae family in the SIB and TDC groups than in ASD, echoing many reports of lower Lachnospiraceae abundance levels in autistic individuals than TDC [86, 92, 93]. One study reported an increased abundance of the Lachnospiraceae family and significant alterations in fecal and urine metabolites following intervention [94]. Lachnospiraceae are known for their production of short-chain fatty acids like butyrate [95, 96], which serve as key energy sources for intestinal epithelial cells [97] and may influence host neurodevelopment through specific mechanistic pathways [98]. Additionally, they exhibit anti-inflammatory effects [99] and promote the maturation of the immune system by enhancing colonic T regulatory cell populations [100–102]. While these findings align with prior research, inconsistencies and even contradictory results have been reported across studies [103, 104]. Such discrepancies may arise from a combination of inherent heterogeneity of autism, methodological differences, and variations in cultural, geographical, and environmental contexts [105].

In our study, SIB showed elevated abundance in the family Prevotellaceae, specifically in the genus Prevotella and Alloprevotella, without significant differences between ASD and TDC. The Prevotellaceae family encompasses four distinct genera: Prevotella, Alloprevotella, Halella, and Paraprevotella [106]. The significantly higher abundance of SIB might indicate a protective effect on siblings from developing ASD. Several studies have consistently reported lower Prevotella abundance in autistic children compared to SIB or TDC [15, 25, 27, 107, 108]. Prevotella, a commensal species involved in saccharide metabolism and vitamin biosynthesis [9, 109], may be therapeutically beneficial in alleviating autistic symptoms [16]. Prevotella converts tryptophan to indole [110], and its abundance correlates with fecal tryptophan levels [111]. The Alloprevotella genus, producing butyric acid [112], exhibits reduced levels in ASD patients [92, 113] and is involved in tryptophan metabolism, negatively associating with tryptophan levels [114] but positively with tryptophan metabolites [115]. The Alloprevotella genus might foster an anti-inflammatory environment [116], possibly reducing gut inflammation in child siblings.

The results from functional prediction indicated higher levels of glutamate degradation in both ASD and SIB, supporting the growing evidence of an excitatory/inhibitory imbalance, particularly regarding glutamatergic dysfunction, in the etiology of ASD [117–119]. Wang et al. [120] reported modifications in gastrointestinal glutamate metabolism associated with alterations in gut microbiota composition in autistic children. However, these predicted functions are mainly derived from the significance of microbiota, warranting careful interpretation. A comprehensive understanding of the role of glutamate in ASD requires multi-genomic investigations, integrating genetic, metabolomic, and microbiota data.

Our investigation identified reduced Anaerostipes and Blautia abundance levels in ASD compared to TDC, aligning with earlier findings [10]. Furthermore, the genus Anaerostipes showed negative associations with social impairment, internalizing problems, and emotional dysregulation. These findings align with a study linking lower plasma IL-6 levels to the relative presence of butyrate-producing bacteria, including Anaerostipes and Coprococcus, in severely autistic individuals [121]. Similarly, a notable inverse relationship between Anaerostipes and plasma IFN-γ levels in severe ASD cases was also found [121]. Majerczyk et al. [122] suggested the role of IL-6 and IFN-γ in facilitating maternal immune responses, potentially leading to gestational neuroinflammation and the development of autistic behaviors during gestation. These findings and our results suggest that imbalances in the Lachnospiraceae family of bacteria, particularly Anaerostipes, are pivotal characteristics of gut microbiota dysbiosis associated with ASD.

The genus Anaerostipes, a Lachnospiraceae member, is a significant butyrate producer in humans and animals [109], metabolizing carbohydrates, lactate, and acetate into butyrate [123]. The latest review also shows the broader role of the gut microbiome in sociability, highlighting how microbial metabolites like butyrate can influence oxytocin levels, a hormone central to social bonding [124]. Reduced Anaerostipes are observed in patients with inflammatory bowel diseases, irritable bowel syndrome [125], metabolic disorders [126], Clostridioides Difficile infection [127], and food allergies [128]. This imbalance could explain the heightened occurrence of functional gastrointestinal issues in autistic individuals. Anaerostipes levels are reduced in MDD, potentially promoting inflammation and oxidative stress [129], which may disrupt brain regions involved in social cognition and contribute to social deficits. Additionally, our findings suggest that gut microbiota dysbiosis may be a predictive marker of ASD severity and emotional/behavioral problems, as evidenced by the inverse relationship between total SRS scores and CBCL scores and the abundance of Anaerostipes. Future research should investigate the causal role of Anaerostipes-derived metabolites in modulating brain function and behavior using animal models and interventional studies such as prebiotic or probiotic supplementation aimed at restoring Anaerostipes levels.

When interpreting our findings, some limitations should be acknowledged. First, the varying sex/gender ratios observed among the ASD, SIB, and TDC groups could introduce potential confounding factors. The small number of female participants also limits our ability to conduct subgroup analyses to assess whether gender moderates or affects the psychometric properties. To address the sex discrepancy, we have accounted for sex as a covariate in our association analyses between microbiota abundance and clinical features. Second, the strict inclusion/exclusion criteria, excluding those with intellectual disabilities or co-occurring psychiatric conditions despite the high comorbid conditions in autistic individuals [130], call for further investigation in a more representative sample of autistic populations. Third, it is noteworthy that the autistic participants displayed a lower FIQ compared to their SIB and TDC participants. Consequently, we have incorporated FIQ as a covariate in our analyses to mitigate potential confounding effects. Fourth, the study’s cross-sectional design limits our ability to determine causal relationships between ASD diagnoses and microbial profiles, or to establish the direction of these associations. Furthermore, longitudinal studies must elucidate the temporal sequence and potential causal pathways between the microbiota and clinical presentation. Fifth, the broad age range (4–25 years) and limited statistical power impede age-stratified analyses, making it difficult to account for developmental factors in lifestyle and dietary habits adequately. At the same time, the Taiwanese national nutrition intake survey suggests that dietary patterns and nutritional intake do not exhibit substantial variations across different age groups [131]. This finding supports our approach to incorporating both children and adults into the current analysis. Furthermore, we included age and upper GI symptoms as covariates in the linear regression models, and they did not significantly influence the alpha diversity results. Sixth, LEfSe analysis may be influenced by zero-inflation in relative abundance data, potentially impacting the accuracy of biomarker identification, and it does not support the inclusion of confounding variables in the analysis. More advanced methods should be considered in future analyses. Seventh, PICRUSt2-based functional prediction from 16S rRNA data may not accurately capture microbial functional potential. Future investigations incorporating metagenomic or metabolomic analyses are warranted. Eighth, the results from the Mantel test should be interpreted with caution due to the relatively low rho value. Lastly, we should be cautious when interpreting results from species-level analysis using shorter 16S amplicons. Ninth, the K-SADS-E and CBCL were used for participants outside their validated age ranges (6–18 years) to maintain consistency in assessment across our samples, which may introduce potential biases in psychometric properties for older participants. Future studies employing long-read sequencing techniques are likely to enhance taxonomic resolution at the species and strain levels.

Conclusion

Our study uncovers distinct taxonomic diversities and microbial compositions in autistic individuals and their SIB. Notably, with various independent contrast methods, we found that Prevotellaceae at the family level displayed increased abundance in SIB. Lachnospiraceae at the family level showed increased abundance in both SIB and TDC groups, suggesting a potential protective role against the development of ASD. Furthermore, we observed that a higher relative abundance of Anaerostipes, a member of the Lachnospiraceae family, correlated with less severe autistic symptoms and fewer emotional and behavioral problems, implying its potential importance in ASD. The microbiota’s involvement in glutamate metabolism emerges as a promising area for the future development of ASD treatment.

Author contributions

Jung-Chi Chang: Conceptualization, Formal analysis, Investigation, Writing – original draft preparation, and Visualization; Yu-Chieh Chen: Conceptualization, Methodology; Hai-Ti Lin: Conceptualization, Methodology, Results evaluation and interpretation; Yan-Lin Chen: Conceptualization and Methodology; Susan Shur-Fen Gau: Principal Investigator, Study design, Conceptualization, Project administration, Data curation, Supervision, Funding acquisition, Results evaluation and interpretation, and Critical revision of the manuscript for submission. The authors have read and approved the final manuscript.

Data availability

The data that has been used is confidential.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41398-025-03768-8.