Microbiome-Specific Statistical Modeling Identifies Interplay Between Gastrointestinal Microbiome and Neurobehavioral Outcomes in Patients With Autism: A Case Control Study

Abstract

Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder with unclear mechanisms of pathogenesis. Gastrointestinal microbiome alterations were found to correlate with ASD core symptoms, but its specific role in ASD pathogenesis has not been determined. In this study, we used a case-control strategy that simultaneously compared the ASD gastrointestinal microbiome with that from age-sex matched controls and first-degree relative controls, using a statistical framework accounting for confounders such as age. Enterobacteriaceae (including Escherichia/Shigella) and Phyllobacterium were significantly enriched in the ASD group, with their relative abundances all following a pattern of ASD > first degree relative control > healthy control, consistent with our hypothesis of living environment and shared microbial and immunological exposures as key drivers of ASD gastrointestinal microbiome dysbiosis. Using multivariable omnibus testing, we identified clinical factors including ADOS scores, dietary habits, and gastrointestinal symptoms that covary with overall microbiome structure within the ASD cohort. A microbiome-specific multivariate modeling approach (MaAsLin2) demonstrated microbial taxa, such as Lachnoclostridium and Tyzzerella, are significantly associated with ASD core symptoms measured by ADOS. Finally, we identified alterations in predicted biological functions, including tryptophan and tyrosine biosynthesis/metabolism potentially relevant to the pathophysiology of the gut-brain-axis. Overall, our results identified gastrointestinal microbiome signature changes in patients with ASD, highlighted associations between gastrointestinal microbiome and clinical characteristics related to the gut-brain axis and identified contributors to the heterogeneity of gastrointestinal microbiome within the ASD population.

Keywords: autism spectrum disorder; biological pathway; gut microbiome; gut-brain axis; multivariable omnibus testing; predictive functional profiling.

Conflict of interest statement

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2021 Huang, Liu, Liu, Chen, Wei, Feng, Wu, Fong, Tian, Wang, Budjan, Zhuang, Wan and Kong.

Figures

Figure 1

Overview of groupwise subject gut microbiome ecological diversity and differentially enriched gut microbiota that are most prevalent in ASD group subjects, followed by healthy controls and first-degree relative controls, respectively. (A) Groupwise comparison of alpha diversity determined via Shannon index in all groups. (B) Alpha diversity via the Shannon index displays significant inverse correlation with ADOS overall score. (C) PCoA of species after filtering (Bray-Curtis) with 95% confidence ellipses. ASD microbiota showed (D) a lower abundance of Akkermansia but a higher abundance of (E) enterobacteriaceae, (F)phyllobacterium, (G)E. Shigella, and (H)E. coli when compared to first-degree relative controls and healthy controls.

Figure 2

The ASD gut microbiota displayed significant differences in predicted metabolic pathways when compared to healthy controls. (A) ASD microbiota expressed lower levels of D-Gln, D-Glu metabolism and higher levels of (B) Trp, (C) Tyr, and (D) fatty acid metabolism. (E) ASD microbiota displayed lower levels of Phe, Try, and Trp biosynthesis and (F) higher levels of Lys degradation. Significant indices were chosen at a Q-value cutoff of 0.1.

Figure 3

ASD severity was associated with gut microbiota abundance. (A) Stereotypic behavior was positively correlated to Tyzzerella relative abundance. (B–D,F–H) ADOS subscale scores (social affect, SA), total scores, and standard score (CSS) were correlated with relative abundance of bacterial taxa. (E) GI severity index score was positively associated with Butyricicoccus abundance. Indices were chosen at a univariate Q-value (FDR-adjusted P-value) cutoff of 0.1.