Mixed Evidence for Impact of Early Infant Gut Microbiome and Later Development of Autism Spectrum Disorder in the MARBLES Prospective Cohort Study

ABSTRACT

This study investigated the relationship between early infant gut microbiome composition and subsequent neurodevelopmental outcomes. Fecal samples from children in the markers of autism risks in babies‐learning early signs (MARBLES) study, a cohort with elevated likelihood of autism, were collected between 0 and 7 months of age and analyzed using 16S rRNA sequencing to evaluate whether the gut microbial composition during early infancy is associated with later neurodevelopmental diagnoses. Clinical classification as autism spectrum disorder (ASD), non‐typically developing without ASD (non‐TD), or typically developing (TD) was completed around 36 months of age using gold‐standard assessment tools. Overall, no significant differences in alpha diversity or beta diversity, nor any differentially abundant bacterial taxa, were found between groups of infants who developed ASD or non‐TD compared to those who went on to have TD. Nonetheless, our findings highlight some early differences in gut microbial composition during infancy that may relate to later neurodevelopmental outcomes. Before adjusting for multiple comparisons, infants who later developed ASD had slightly lower levels of Veillonella and Flavonifractor genera compared to children who were later found to be TD. These results suggest specific bacterial taxa may already differentiate in early infancy, but may be more subtle than other factors, such as mode of delivery and diet during early infancy. To understand longitudinal trajectories of the gut microbiome in association with later neurodevelopment, future studies should include a larger cohort to detect smaller effect sizes or investigate later time points in infancy.

Lay Summary

The human gut has many types of bacteria, some beneficial and some that may be harmful to health. Together these bacteria are referred to as the gut microbiome. Previous research shows that the bacteria in the human gut may differ in older children diagnosed with autism spectrum disorder compared to typically developing children. Children with autism are known to have problems with their gastrointestinal (GI) tract, though the relationship between GI health and autism is not well understood, particularly how the bacteria in the gut change in early life, before ASD behaviors begin to be seen. This study focuses on the gut microbiome in the first 7 months after birth and its relation to later development of autism, examined at 36 months of age. We found no major differences in the gut microbiome of infants who developed autism compared to those who did not. However, infants who later developed autism had slightly fewer of two bacteria types (Veillonella and Flavonifractor) early in life compared to infants who developed typically. However, these differences are minor compared to other factors, like whether the baby was born by C‐section or what they ate, which might matter more for the gut microbiome composition at this age than their later developmental diagnosis.

1. Introduction

Autism spectrum disorder (ASD) is a neurodevelopmental condition that is currently estimated to affect 1 in 31 children in the US (Shaw et al. 2025). Although ASD is defined by difficulties in social communication and the presence of restricted interests or repetitive behaviors, a range of co‐occurring conditions such as immune dysregulation (Ashwood et al. 2006; Onore et al. 2012), metabolic abnormalities (Adams et al. 2011; Orozco et al. 2019; Shen et al. 2020), gastrointestinal (GI) symptoms (Chaidez et al. 2014; Hsiao 2014), and gut microbial dysbiosis (Saurman et al. 2020; Vuong and Hsiao 2017) are frequently reported in individuals with autism. The gut microbiome, in particular, has been associated with the pathogenesis of many neurological conditions, including autism (Ahrens et al. 2024; Mulle et al. 2013), and may be implicated in immune dysregulation, metabolic disturbances, and GI issues in ASD (Ahrens et al. 2024; Rose et al. 2018).

Several cross‐sectional studies have investigated the gut microbiome in association with ASD (Coretti et al. 2018; Dan et al. 2020; Ding et al. 2021; Finegold et al. 2010; Ma et al. 2019; Parracho et al. 2005; Wang et al. 2014). Although the specific genera of microbes altered in ASD cases versus controls have varied between studies, three meta‐analyses found some overlap in particular species, including a lower abundance of Bacteroides and Bifidobacterium among individuals with ASD, as well as a higher abundance of Clostridium and Faecalibacterium—this highlights how alterations in particular species have repeatedly been associated with ASD (Andreo‐Martinez et al. 2022; Iglesias‐Vazquez et al. 2020; Xu et al. 2019). However, previous studies have primarily focused on older children with a confirmed diagnosis, which occurs after the microbiome has begun to establish and undergone composition alterations. Therefore, the timing of the onset of modifications to the early infant gut microbiome has not been thoroughly investigated. Most recently, a Swedish cohort study was able to leverage stool samples collected at age 1 year to explore microbiome species that differed in several neurodevelopmental conditions, including ASD. This study found differences in the abundance of several microbes associated with autism, including lower levels of Akkermansia muciniphila in the samples from children who later developed autism (Ahrens et al. 2024). As the gut microbiome is heavily influenced by diet (David et al. 2014), and as ASD symptoms only rarely begin to emerge around the first birthday (Maestro et al. 2005), even the study by Aherns et al. is unable to determine whether the gut microbiome disturbance observed with ASD occurs before behavioral issues or, alternatively, whether the ASD behaviors may influence both diet and the gut microbiome.

Understanding the gut microbiota and their trajectories in early life could provide insight into potential impacts on ASD and help determine whether microbiome alterations influence ASD development or if symptoms associated with ASD lead to changes in microbiome composition. The infant's developing gut microbiome is influenced by early life events, including the mode of delivery (cesarean section versus vaginal delivery), infant diet type (e.g., breastfeeding or formula feeding), and the duration of breastfeeding (Fallani et al. 2010; Ho et al. 2018). Just 3 weeks after birth, the metabolic products of gut flora have been shown to predict an increased risk of allergies or asthma in later life (Fujimura et al. 2016; Valverde‐Molina and Garcia‐Marcos 2023). The mix of immune and gut‐related symptoms observed in ASD underscores the question of how early microbiome changes associated with ASD occur and if these changes predate the onset of symptoms. This study explores the infant gut microbiome to identify whether specific changes in the microbiota during early infancy are associated with later diagnosis of ASD, addressing a gap in understanding the early onset and mechanisms of gut microbial changes in individuals with ASD.

2. Methods

2.1. Study Participants

All participants in the present study are from the MARBLES (Markers of Autism Risk in Babies—Learning Early Signs) Study, an enriched‐risk longitudinal cohort that began recruiting participants in 2006. Participating mothers had a previous child with ASD and therefore were at increased risk (~20%) for delivering another infant who would develop ASD (Ozonoff et al. 2024). Early‐life stool samples from the infants were collected beginning for participants enrolled after 2011. We attempted to collect longitudinal stool samples from children every 2 weeks, starting at 2–12 weeks (3 months) and then at 4, 5, 6, 9, 12, 24, and 36 months of age. Stool samples were collected by parents into a plastic container or from a soiled diaper. Parents were given a large conical tube and a wooden stick and instructed to scoop the sample from the diaper or baby's bottom and put it in the provided tube. The samples were frozen immediately at −20°C in a home freezer before being transported to the lab and stored at −80°C. For the present study, we focused on stool samples collected between 0 and 7 months of age to focus on the gut microbiome in early infancy. The details of the MARBLES Study have previously been published (Hertz‐Picciotto et al. 2018). Briefly, at 36 months of age, children in the MARBLES study were assessed for ASD using the Autism Diagnostic Observation Schedules (ADOS) (Lord et al. 2012) and the Mullen Scale of Early Learning (MSEL) (Mullen 1995) as previously described (Ozonoff et al. 2014; Schmidt et al. 2021). However, for three participants in our study, the Mullens and/or ADOS were conducted at 43 months of age or older. Children who received a clinical best estimate diagnosis of ASD by a trained psychometrician based on Diagnostic and Statistical Manual of Mental Disorders Fifth Edition (DSM‐5) criteria and had ADOS scores above the ASD cutoff were classified as ASD. Children without ASD but who had elevated ADOS scores (within three points of the ASD cutoff) and/or low MSEL subscale scores (i.e., two or more subscale scores greater than 1.5 standard deviations (SD) below average, or at least one subscale more than two SDs below average) were classified as non‐typically developing (non‐TD). Children who did not meet the criteria for ASD or non‐TD were classified as typically developing (TD).

Of the 155 participants in the present analysis, 106 completed the MARBLES study protocol and were assigned a final clinical outcome (ASD = 27, non‐TD = 12, and TD = 67). However, due to COVID‐19 pandemic restrictions, 49 participants did not receive an outcome classification using standard MARBLES procedures. For 31 of the 49 participants, alternative diagnostic tools were used, including the Brief Observation of Symptoms of Autism (BOSA (Dow et al. 2022); n = 28), the TELE‐ASD‐PEDS (TAP (Corona et al. 2021); n = 2), and modified ADOS (with masking; n = 1). Once COVID‐19 restrictions ended, the MARBLES study resumed standardized diagnostic testing, with the exception of three participants who were evaluated at older ages (43 months or more). Additionally, neurodevelopmental outcomes could not be evaluated for 18 children who were lost to follow‐up and excluded from further analysis. Details of the alternative diagnostic instruments are presented in the Supporting Information. Adding these 31 participants, the final analyzed sample included 137 participants with an outcome classification (ASD, n = 42; non‐TD, n = 16; TD, n = 79), corresponding to 465 fecal samples collected between 0 and 7 months of age. Figure S1 shows the distribution of infant fecal samples by collection time and neurodevelopmental outcome for the study participants. MARBLES was approved by the Institutional Review Board at the University of California, Davis, and by the State of California Department of Developmental Services, and informed consent was obtained from all participants before enrollment.

2.2. Fecal Microbiome Analysis

The fecal microbiome was assessed for this study using 16S rRNA gene sequencing. DNA was extracted from approximately 200 mg of feces that were aliquoted from the infant specimens. DNA was extracted using the KingFisher Flex (Thermo Fisher Scientific, Waltham, MA, USA) and the ZYMOBiomics 96 MagBead DNA kit (Zymo Research) and included a bead beating step as previously described (Melendez Hebib et al. 2023). The V4 region of the 16S rRNA gene was amplified and sequenced as described in Huda et al. (2014). The resulting reads were processed using the open‐source R package Divisive Amplicon denoising algorithm (DADA2) (Callahan et al. 2016) as implemented in the bioinformatics pipeline Quantitative Insight Into Microbial Ecology QIIME2‐2020.2 (Bolyen et al. 2019). Amplicon sequence variants (ASVs) were assigned taxonomy using the prebuilt SILVA classifier (Quast et al. 2013). Due to the wide variation in read depth, samples were rarefied to 2556 reads per sample before analysis.

2.3. Statistical Analysis

To investigate differences in the fecal microbiome, we analyzed two alpha diversity metrics: the Shannon index (Shannon 1948) and the observed abundance of ASVs per sample. Differences in Shannon diversity were examined using a random intercept linear mixed‐effects model, adjusting for covariates specified in the directed acyclic graph (DAG; detailed below), with subject ID included as a random term in the lmerTest R package (Bates, Mächler, et al. 2015a). For microbial richness, we used Poisson generalized linear mixed models (as observed ASV is a count measure of richness), adjusting for confounders identified in the DAG, with subject ID as a random effect in the lme4 R package (Bates, Maechler, et al. 2015b). Both models included an interaction term between neurodevelopmental outcome group and age. Beta diversity was assessed using the robust Aitchison distance. Differences in beta diversity were visualized using nonmetric multidimensional scaling (NMDS). Scree plots were generated to select the number of axes needed for the NMDS plots, which resulted in four dimensions being selected. PERMANOVA models were run on the data separately for each stratum defined by three different age groups (0–1 month, 2–3 months, and 4–7 months) to investigate beta‐diversity differences in early infancy, adjusting for covariates defined in the DAG. Because only one sample (the earliest one) per infant was used for each age group in the PERMANOVA models to avoid issues with repeated measures, the child's age was not entered as a covariate into the models (as it was inherently controlled through stratification); this included 105 subjects with samples between 0 and 1 month of age, 87 subjects with samples between 2 and 3 months, and 92 subjects with samples between 4 and 7 months of age. We used Analysis of the Composition of Microbiomes with Bias Correction 2 (ANCOM‐BC2; Lin and Peddada 2020) with a random intercept for subject ID to model repeated measures to evaluate whether specific taxa (at the ASV and family/genus levels) were associated with the outcome group (ASD or non‐TD compared to TD controls) adjusting for covariates as defined in the DAG. The effect size in ANCOM‐BC2 is represented as a log‐fold change (LFC). An interaction term for neurodevelopmental outcome and age was tested to evaluate if differential abundance differed by group over time in a separate model. ANCOM‐BC2 results were adjusted for multiple testing using the false‐discovery rate (FDR), where q < 0.05 was considered statistically significant (q‐values are FDR‐adjusted p‐values).

All models were adjusted for covariates selected a priori based on a DAG (Figure S2). Possible covariates considered in our DAG included: infant feeding throughout the entire first 6 months of life (categorical: exclusive breastfeeding (BF) or mixed feeding), infant feeding at the time of fecal sample collection (categorical: (a) BF only, no food; (b) formula only, no food; (c) BF and formula, no food; (d) formula only and food; (e) BF + food), infant biological sex (categorical: male or female), delivery mode (categorical: vaginal or c‐section), child's age at sample collection (continuous: days), and parental homeownership (categorical: homeowner or renter), as an indicator of socioeconomic status and family wealth. Because the two infant feeding variables we considered in our DAG model (i.e., infant feeding throughout the first 6 months vs. infant feeding at the time of sample) were highly correlated, and because we strove for the most parsimonious model while still adjusting for confounders, we selected to retain the infant feeding throughout the first 6 months of life, to represent feeding during early infancy. As such, the final DAG‐based model included infant feeding throughout the first 6 months of life, parental homeownership, mode of delivery, and child's age, which were considered for adjustment in further analysis. All statistical analyses were conducted in R (version 4.4.0).

3. Results

Table 1 summarizes the characteristics of the 137 participants in the present study. Fifty‐seven percent of our study participants were male compared to 43% female. Approximately 64% of study participants were born vaginally, and 36% were born via cesarean. All infants received some breastmilk during the study period, with approximately 43% exclusively breastfed and 58% receiving a combination of breastmilk and formula.

TABLE 1.

Characteristics of study participants.

Characteristic a	Overall (N = 137)
Neurodevelopmental diagnosis
TD	79 (57.7%)
ASD	42 (30.6%)
Non‐TD	16 (11.7%)
Sex
Female	59 (43.1%)
Male	78 (56.9%)
Mode of delivery
Cesarean	49 (36.3%)
Vaginal	86 (63.7%)
Infant feeding in the first 6 months of life
Mixed feeding (breastfed and formula)	79 (57.7%)
Breastfeed exclusively	58 (42.3%)
Parental homeownership
Owner	75 (56.8%)
Renter	57 (43.2%)

^a Missing (n): Mode of delivery (2), parental homeownership (5).

3.1. Alpha and Beta Diversity Analyses

In the models adjusted for potential confounders, children who went on to develop ASD or non‐TD did not have significantly different alpha diversity (Shannon's diversity index) or microbial richness (defined by the observed ASVs) in their infant fecal microbiomes compared to children who went on to have TD (Figure 1). When we examined the alpha diversity across other covariates, we found that the child's mode of delivery was not significantly correlated with differences in Shannon diversity or the number of observed ASVs. However, children who received mixed feeding (breastmilk and formula) had higher Shannon diversity (p = 0.002) and greater microbial richness (p = 0.001) than exclusively breastfed infants. Shannon diversity (p = 0.007) and observed ASVs (p = 0.000) increased with age; however, no significant interaction term was found between neurodevelopmental outcomes and age, indicating that alpha diversity did not change differentially by age across the neurodevelopmental outcomes. Figure S3 shows longitudinal changes in alpha diversity across age.

FIGURE 1.

Alpha diversity as measured by (A) Shannon's diversity index and (B) observed ASVs by neurodevelopmental outcomes, mixed feeding (i.e., breastfed/formula‐fed infants) vs. exclusively breastfed infants, mode of delivery, and age.

Beta diversity was measured using robust Aitchison distance to assess dissimilarity between samples based on neurodevelopmental outcomes. Figure 2 shows nonmetric multidimensional scaling (NMDS) plots based on the robust Aitchison's distance across the different time points by neurodevelopmental outcomes. Similarly to alpha diversity, there were no significant differences in beta diversity between TD, non‐TD, and ASD participants at any of the three stratified time points in infancy (0–1 months of age, p = 0.202; 2–3 months, p = 0.391; and 4–7 months, p = 0.107) based on PERMANOVA models. At 0–1 month of age, we found that microbial communities differed when comparing infants who received mixed feeding with those who were exclusively breastfed during the first 6 months of life, according to a robust Aitchison distance (p = 0.028). Also, at 0–1 month of age, significant differences in microbial communities were observed when comparing vaginally versus c‐section‐delivered infants (p = 0.001). Likewise, at 4–7 months of age, significant differences were again observed by type of feeding (p = 0.004) and by mode of delivery (p = 0.021). Figure S4 shows NMDS plots colored by infant feeding and delivery mode across the three time points.

FIGURE 2.

Nonmetric multidimensional scaling (NMDS) plots based on the robust Aitchison's distance, points are colored based on TD, ASD, and non‐TD categories: (A) 0–1 month samples, (B) 2–3 month samples, and (C) 4–7 month samples.

3.2. Differential Abundance

To evaluate potential links between specific taxa and neurodevelopmental outcomes, we implemented differential abundance analysis using ANCOM‐BC2, adjusting for infant feeding in the first 6 months of life, mode of delivery, parental homeownership, and child's age. Using ANCOM‐BC2 in models adjusted for multiple comparisons, we did not find any significant ASV/family/genus that were differentially abundant when comparing ASD and non‐TD children with those who went on to have TD. We also did not find a significant interaction between neurodevelopmental outcome group and age. However, our ANCOM‐BC2 analysis revealed statistically significant age‐related changes in microbial abundance (Figure S5), which included an increase in the relative abundance of Fusicatenibacter (log‐fold change [LFC]: 0.020, q = 0.002), [Ruminococcus]_gnavus_group (LFC: 0.014, q = 0.000), Blautia (LFC: 0.013, q = 0.001), Erysipelatoclostridium (LFC: 0.011, q = 0.000), and Bifidobacterium (LFC: 0.011, q = 0.000) per day of age. As well as the decrease in the relative abundance of Staphylococcus_1 (LFC: −0.021, q = 0.000), Staphylococcus_2 (LFC: −0.027, q = 0.000), Streptococcus (LFC: −0.009, q = 0.000), and Escherichia‐Shigella (LFC: −0.008, q = 0.002) at the genus level, and Enterobacteriaceae_1 (LFC: −0.014, q = 0.000), and Enterobacteriaceae_2 (LFC: −0.007, q = 0.023) at the family level per day of age.

Because our primary outcome of interest was neurodevelopmental outcome, it is worth noting that before adjusting for multiple comparisons, we observed that infants who developed ASD showed slightly lower levels of Veillonella (LFC: −0.85, p = 0.07, q = 0.98) and Flavonifractor (LFC: −1.29, p = 0.02, q = 0.98) at the genus level compared with infants who were TD based on ANCOM‐BC2 results (Figure S6). Similarly, infants who later developed non‐TD appeared to have modestly higher relative abundance of Enterococcus (LFC: 1.86, p = 0.003, q = 0.96) and [Ruminococcus]_gnavus_group ( R. gnavus ) (LFC: 1.06, p = 0.08, q = 0.96) in early infancy compared to those with TD, but further research is warranted. Results from ANCOM‐BC2 models are presented in Table S1. The relative abundance of the most common families detected in the samples compared across later neurodevelopmental outcomes for each age group is shown in Figure 3.

FIGURE 3.

Relative abundance of the nine most common families at the three age ranges (0–1, 2–3, and 4–7 months). For clarity, the subset of only one sample per infant per time point used in the beta‐diversity analysis is displayed here, and the x‐axis is labeled by later neurodevelopmental outcomes.

4. Discussion

Several studies have reported gut microbiome differences among children with ASD compared to neurotypical controls (Coretti et al. 2018; Dan et al. 2020; Ding et al. 2021; Finegold et al. 2010; Ma et al. 2019; Parracho et al. 2005; Wang et al. 2014). However, the existing literature has primarily focused on older children with a confirmed ASD diagnosis—therefore, the timing of when shifts in gut microbial dysbiosis initiate in children with ASD remains unknown. Given the increased understanding of the gut‐brain axis, there is a possibility that microbial shifts associated with ASD occur prior to diagnosis. As such, the present study aimed to investigate gut microbial differences in early infancy, specifically focusing on the first months of life, prior to diagnosis and a period of dynamic changes to the infant gut microbiome. This study sought to capture the onset of early shifts in the gut microbiome compositions associated with adverse neurodevelopment.

Overall, we found no significant differences in alpha diversity or beta diversity, and no bacterial taxa were differentially abundant between groups of infants who were later classified as having ASD or non‐TD compared to those classified as TD. Although others have found significantly lower alpha diversity among children with ASD compared to neurotypical controls (Ding et al. 2021; Ma et al. 2019; Zhao et al. 2023), these discrepancies may reflect differences in study populations (such as our investigation of a population with high likelihood of developing ASD vs. general populations), the heterogeneity of ASD, and differences in ages investigated (early infancy vs. older children), which may contribute to these discrepancies. On the other hand, our methods were able to detect differences across other factors known to be associated with variations in the early infant gut microbiome. Specifically, we found that infants who received mixed feeding (a combination of formula and breastmilk) had significantly higher alpha diversity compared to exclusively breastfed infants during the first 6 months of life. Similarly, a recent study on infants at 4–5 months found that exclusively formula‐fed infants had a greater alpha diversity than exclusively breastfed infants (Odiase et al. 2023). A meta‐analysis also concluded that nonexclusive breastfeeding is associated with an increased alpha diversity before 6 months compared with exclusive breastfeeding (Ho et al. 2018). We also found that alpha diversity increased with age, which aligns with previous reports of increased richness of the microbial community as children grow older (Roswall et al. 2021). In addition, PERMANOVA analysis of beta diversity revealed distinct fecal microbiota in infants born via cesarean section compared to those born vaginally, particularly in samples from 0 to 1 month of age. Indeed, the mode of delivery has been recognized as a significant determinant of microbiota composition through the first years of life (Fallani et al. 2010; Ho et al. 2018; Wernroth et al. 2022). As such, there are a few possibilities to consider. First, we may need a larger study to detect subtle differences in the microbiome associated with later neurodevelopmental outcomes. Second, the shift in the gut microbiome associated with ASD and non‐TD may occur at ages older than 7 months, as other studies have identified differences as early as 12 months (Ahrens et al. 2024). If so, the early complementary feeding period may be critical to understanding the developing gut microbiome in relation to neurodevelopmental conditions. However, given that diet significantly impacts the gut microbiome composition and that eating problems—like severe picky eating, restrictive diets, and aversion to new/unknown foods—are common among individuals with ASD (Ledford and Gast 2006), it is possible that this behavioral tendency could contribute to some of the gut microbiome differences reported in older children with ASD (Baraskewich et al. 2021), a pathway that should be considered.

Although the primary focus of our investigation was the infant gut microbiome as a function of neurodevelopmental outcomes, examining how early life factors (such as age, environmental exposures, and diet) affect the infant gut microbiome is critical for understanding its developmental trajectory. In the first 7 months of life, we found longitudinal shifts in the infant gut microbiome from facultative anaerobes like Staphylococcus and Streptococcus to obligate anaerobes such as Fusicatenibacter, R. gnavus , Blautia, Erysipelatoclostridium, and Bifidobacterium, reflecting a transition to an anaerobic environment and a shift from lactic acid production (primarily by Bifidobacterium) to increased short‐chain fatty acid (SCFA) production. SCFAs, like butyrate (e.g., produced by Blautia or R. gnavus ), are beneficial metabolites supporting gut barrier function and immune regulation, and serve as an energy source for intestinal epithelial cells (Baxter et al. 2019; Mann et al. 2024; Takada et al. 2013). Additionally, dietary fiber, particularly human milk oligosaccharides from breast milk and later complex carbohydrates from complementary foods, provide fuel for SCFA‐producing bacteria, allowing them to thrive and increasing with microbiota maturation (Cronin et al. 2021; Crost et al. 2023). These results indicate that significant changes occur in the gut microbiota composition as the infants mature, partly due to environmental factors (such as oxygen availability, diet, and mode of delivery). Our results align with previous studies, which highlight significant shifts in microbiota composition during infancy, with the strongest changes reported occurring between the ages of 4 and 12 months (Roswall et al. 2021).

While we observed no significant differences in bacterial abundance in early infancy associated with later neurodevelopmental outcomes, our findings may provide clues of subtle microbial changes in early infancy that align with existing literature on older children, potentially offering insight into early dysbiosis. For example, before FDR correction, we found evidence of slightly differentially abundant gut bacteria among infants with subsequent ASD diagnoses (such as decreased Veillonella) compared to those later classified as typically developing. Though these results were not significant after correcting for multiple comparisons, they align with previous studies conducted in older children/teens with confirmed ASD diagnosis, including work by Strati et al. that similarly reported a decreased abundance of Veillonella genus in an ASD cohort (average age 11.1 ± 6.8) (Strati et al. 2017) as well as Kang et al. who also noted lower abundance in autistic children (3–16 years old) compared to neurotypical controls (Kang et al. 2013). Interestingly, Veillonella has been shown to promote the production of beneficial SCFA metabolites (Macfabe 2012; Zhang and Huang 2023), some of which were previously shown to be lower in ASD individuals (4.43 ± 1.47 years old) (Liu et al. 2019)—although others have also reported elevated SCFA levels in ASD (4–10 years of age) (De Angelis et al. 2013). In our study, children who went on to have ASD had a modestly lower relative abundance of Flavonifractor genus between 0 and 7 months compared to children who were classified as TD. Zou et al. previously also reported that Flavonifractor was decreased in children (2–7 years old) with confirmed ASD diagnosis compared to controls (Zou et al. 2020), as did Ma and colleagues among children (6–9 years of age) with ASD compared to sex‐matched controls (Ma et al. 2019). However, some differences may arise at the species level, as others have reported that Flavonifractor plautii species were significantly more abundant in ASD microbiota than in neurotypical controls in a study that included children aged 3–18 years (Luna et al. 2017).

Furthermore, before adjusting for multiple comparisons, when we compared the early infant microbiome among children who went on to develop nontypically but without ASD (non‐TD) versus those who were classified as TD, we found modestly elevated levels of R. gnavus and Enterococcus. This suggests the early infant gut microbial colonization of children with non‐TD differed from both those who were later classified as TD and those with ASD. Although R. gnavus is part of a healthy human gut microbiota and known SCFA‐producing bacteria (as previously mentioned), several studies have shown a positive correlation between R. gnavus and gut‐related diseases (Crost et al. 2023), including Crohn's disease (Feng et al. 2022; Joossens et al. 2011), ulcerative colitis (Shin et al. 2023), inflammatory bowel disease (Willing et al. 2010), and irritable bowel syndrome (Han et al. 2022; Jeffery et al. 2020). Additionally, an increased abundance of R. gnavus was found in patients suffering from general anxiety disorders compared to healthy individuals (Jiang et al. 2018), and positive correlations have been reported between R. gnavus and epilepsy (Dong et al. 2022) and patients with Parkinson's disease (Lubomski et al. 2022). Similarly, Enterococcus are mainly commensal organisms that colonize the digestive system of infants in the first days after birth, and some Enterococcus strains are even used as probiotics (Daca and Jarzembowski 2024; Krawczyk et al. 2021). However, like other bacteria, disturbances in the balance of specific microbiota can lead to dysbiosis, and Enterococci have also been implicated in urinary tract infections, bloodstream infections, endocarditis, and other infections (Franz et al. 1999; Rosselli Del Turco et al. 2021). Overall, our results investigating the infant gut microbiome in association with the later development of ASD and non‐TD may have been underpowered to detect modest associations between the early infant gut microbiome and later neurodevelopmental outcomes within this time frame, when the gut microbiome is in its most dynamic stage. Indeed, gut microbiome alterations due to neurodevelopmental outcomes may be more subtle during early infancy, at least in comparison with factors such as delivery mode and feeding practice.

Our study addresses a critical gap in the literature by prospectively investigating the gut microbiome of infants before neurodevelopmental diagnoses were obtained, therefore adding an essential contribution to the limited knowledge about early gut microbial differences in children who develop ASD. We did not find that the early infant gut microbial composition at 0–7 months among children who developed ASD later significantly differed compared to children who went on to develop typically. However, a limitation of our study is that we did not investigate later time points (i.e., 8–12 months) in infancy. Our results do not rule out the possibility that gut microbial dysbiosis is associated with the later development of ASD, considering that the window of vulnerability may occur later in the first year or the early part of the second year of life. Furthermore, all children were at a higher likelihood of developing ASD and other neurodevelopmental conditions than the general population, which may reflect genetic predisposition playing a larger role than for many children without a family history of ASD. As a result, our findings may not be generalizable to the general population. Additionally, our investigation focused on the bacterial V4 region of the 16S rRNA gene—a hypervariable region routinely used for microbial community analysis and taxonomic classification—but this method does not fully capture species‐ or strain‐level taxonomic differences in microbial composition, some of which may be related to ASD development. A further limitation in our study is that, due to the COVID‐19 pandemic, alternative diagnostic instruments had to be used for a small subset (n = 31) of children in our study. However, the majority (n = 106) were assessed using gold standard instruments and diagnostic procedures. Despite these limitations, a strength of our study is that samples were collected prospectively, allowing us to characterize very early shifts in the infant gut microbiome composition before neurodevelopmental diagnoses were obtained. Our study also detected the impacts of other early life factors, such as mode of delivery, infant feeding, and age, which have a strong influence on the early gut microbiome.

Despite the relatively low power, our uncorrected differential abundance results highlight the potential of early differences in the gut microbiome among children who later developed ASD compared with those who developed typically, for example, the trend of decreased levels of Veillonella among infants who later developed ASD compared to TD controls. Future studies should investigate infants 8–12 months of age—a critical postweaning window when children typically start to consume small amounts of complementary solid foods and the composition of the gut microbiome begins to stabilize and progress toward adult‐like levels (Roswall et al. 2021) and increase the sample size to identify more cryptic and subtle microbiome composition changes. Moreover, with improvement in sequencing reference libraries and with the emergence of metagenomics sequencing, future investigations can obtain a higher taxonomic resolution (particularly below the genus level), aim to investigate all genomic DNA in a sample rather than just one specific region of DNA, and capture additional information about the microbiome's metabolic functional potential. Given that we did not detect a significant difference at the genus level, using metagenomic sequencing for species‐level resolution and the functional capacity of the gut microbiome will be critical for future studies. Although our focus here worked with the premise of a predisposition to ASD having an influence on the gut microbiome, the reverse could also be hypothesized, whereby early life microbiota and its shifts predict later neurodevelopmental outcomes. Understanding longitudinal trajectories of the gut microbiome in association with the later development of ASD could help identify biological mechanisms that predate ASD diagnosis and pinpoint when gut microbial communities shift among a subset of children who will develop ASD. Moreover, the microbiome might function as a mediator of factors such as mode of delivery or early infant diet, both of which have been linked directly to ASD (Liu et al. 2022). Regardless of the directionality of the gut‐neurodevelopmental associations, targeted strategies are needed to prevent or mitigate ASD‐associated gut microbial dysbiosis and to guide the timing of potential intervention strategies—such as fecal transplants (Zhang et al. 2023) and targeted probiotic supplementations (Grimaldi et al. 2018), which could improve ASD symptoms and enable all children to reach their full potential.

Funding

This work was supported by National Institutes of Health (P01ES11269, R01ES020392, R01ES028089, P30ES023513, R/U24ES028533), Eunice Kennedy Shriver National Institute of Child Health and Human Development (P50HD103526), Simons Foundation Autism Research Initiative (SFARI #863967, RJS), U.S. Environmental Protection Agency (RD‐829388, RD‐833292).

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: Distribution (n) of infant ages (from 0 to 7 months) at collection of 465 fecal samples analyzed for microbiome profiles, stratified by child's neurodevelopment outcome at ~36 months.

Figure S2: Directed acyclic graph (DAG) for ASD diagnosis and gut microbiome to identify potential confounders. Green circles represent exposure or ancestors of the exposure, blue circles represent outcome, and pink circles ancestors of both exposure and outcome.

Figure S3: Alpha diversity measured by Shannon's index and observed ASVs across age by neurodevelopmental outcomes. Shannon diversity and observed ASVs both increased with age, but no significant interaction term was found between neurodevelopmental outcomes and age.

Figure S4: Nonmetric multidimensional scaling (NMDS) plots based on the robust Aitchison's distance, points are colored based on (A) infant feeding (mixed feeding vs. exclusively breastfed) and (B) mode of delivery (cesarean vs. vaginal) across 0–1 month samples, 2–3 month samples, and 4–7 month samples.

Figure S5: Differential abundance using Analysis of Composition of Microbiomes with Bias Correction2 (ANCOM‐BC2) shows log fold change (log e) based on age (days) adjusted for infant feeding throughout the entire first 6 months of life, parental homeownership, mode of delivery. Results adjusted for multiple comparisons, with only significant values (q < 0.05) shown.

Figure S6: Differential abundance using Analysis of Composition of Microbiomes with Bias Correction2 (ANCOM‐BC2) shows positive (RED) log‐fold change (log e) (enriched taxa) and negative (PURPLE) log‐fold change (log e) (decreased taxa) comparing ASD and non‐TD to TD controls, adjusted for infant feeding throughout the entire first 6 months of life, parental homeownership, mode of delivery, and child's age. Results are unadjusted for multiple comparisons. Nonsignificant trends (p > 0.1) were replaced by 0.

Table S1: Differential abundance of taxa across ASD and non‐TD groups using ANCOM‐BC2. Results represent the full output of the analysis of compositions of microibome with Bias Correction 2 (ANCOM‐BC2), with TD children as the reference group. Results include: diff_abn, which indicates whether a taxon passed the significance threshold (q < 0.05); lfc, the estimated log‐fold change in absolute abundance; p_val, the raw p‐value; q_val, the adjusted p‐value to control for false discovery rate (FDR); se, the standard error of the LFC estimate; passed_se, which identify taxon that passed the sensitivity analysis; and W, the test statistic. Green shading denotes taxa with significant unadjusted p‐values (p < 0.05) discussed in manuscript.

Data Availability Statement

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