Prospective associations of the infant gut microbiome and microbial function with social behaviors related to autism at age 3 years

The hypothesized link between gut bacteria and autism spectrum disorder (ASD) has been explored through animal models and human studies with microbiome assessment after ASD presentation. We aimed to prospectively characterize the association between the infant/toddler gut microbiome and ASD-related social behaviors at age 3 years. As part of an ongoing birth cohort gut bacterial diversity, structure, taxa, and function at 6 weeks (n = 166), 1 year (n = 158), 2 years (n = 129), and 3 years (n = 140) were quantified with 16S rRNA gene and shotgun metagenomic sequencing (n = 101 six weeks, n = 103 one year). ASD-related social behavior was assessed at age 3 years using Social Responsiveness Scale (SRS-2) T-scores. Covariate-adjusted linear and permutation-based models were implemented. Microbiome structure at 1 year was associated with SRS-2 total T-scores (p = 0.01). Several taxa at 1, 2, and 3 years were associated with SRS-2 performance, including many in the Lachnospiraceae family. Higher relative abundance of Adlercreutzia equolifaciens and Ruminococcus torques at 1 year related to poorer SRS-2 performance. Two functional pathways, l-ornithine and vitamin B6 biosynthesis, were associated with better social skills at 3 years. Our results support potential associations between early-childhood gut microbiome and social behaviors. Future mechanistic studies are warranted to pinpoint sensitive targets for intervention.


New Hampshire Birth Cohort Study details
Recruitment occurred at New Hampshire prenatal clinics at approximately 24-28 weeks gestation. Women were included if they used a private, unregulated well at their residence since their last menstrual period and had no plans to move.

Stool sample processing
Stool samples were stored in participants' home freezer until they were able to return it to the study site (within 24 hours). Stool was then thawed at 4°C so that it could be aliquoted into cryovial tubes containing RNAlater and homogenized before long-term storage at -80°C. DNA was extracted from these stools and both 16S rRNA and metagenomic sequencing were performed.

16S rRNA gene sequencing
Sequencing was performed on an Illumina MiSeq at the Marine Biological Laboratory (MBL) in Woods Hole, Massachusetts 1-3 . Sequences were filtered and trimmed to amplicon sequence variants (ASVs) using the DADA2 pipeline and paired to taxonomies inferred from the Greengenes database 1,4-6 . Due to preliminary ASV-specific results and the hypothesized biological importance of the relative abundance of Bacteroidetes compared to the relative abundance of Firmicutes, this ratio was calculated and used as an outcome in a supplemental analysis 7,8 . Within-subject diversity was assessed using the Shannon 9 and Simpson 10 Indices as well as the count of detected ASVs.

Metagenomic sequencing
Briefly, DNA was sheared to a mean length of 400bp before construction of sequencing libraries following Nugen's Ovation Ultralow V2 protocol and sequencing on an Illumina NextSeq. After trimming and cleaning reads, taxa were identified using MetaPhlAn 11 and pathway relative abundance was inferred using the HUMAnN2 pipeline with default parameters 12 .

Sensitivity analyses
Primary analyses used SRS-2 total T-scores at three years, but associations with raw scores were tested in sensitivity analyses. In other sensitivity analyses, in a subset of participants with available data, we considered third trimester urinary cotinine (comparing detect to non-detect) in place of maternal self-report of smoking. We also conducted an analysis restricted to term infants. Additional sensitivity analyses considered the inclusion of covariates such as an indicator variable for any breastfeeding at six months and a measure of the parent-child relationship (Parenting Relationship Questionnaire-preschool form (PRQ), see below) 13 . The PRQ was included in basic and full models at all time points, whereas six-month breastfeeding was included in basic and full models for the six-week microbiome, but only in full models for the later microbiome ages because it may be a factor for which subsequent microbiome composition acts as a mediator. To explore the possibility that any observed time-specific associations were the result of population-specific effects (i.e., due to selection bias), an analysis was conducted among children with microbiome sequencing available at all time points (n = 21; Supplementary Table S1).

Incorporation of the Parenting Relationship Questionnaire (PRQ)
The preschool version of the PRQ produces five scores assessing different domains of parenting quality including attachment, discipline practices, involvement, parenting confidence, and relational frustration. Each raw score is standardized to a reference population producing Tscores with an average of 50 and a standard deviation of 10. For four of these scales (attachment, discipline practices, involvement, and parenting confidence) a higher score is considered better whereas a higher relational frustration score is considered worse. To incorporate these five scales as one covariate the relational frustration score was transformed as follows: The five scores (attachment, discipline, practices, involvement, parenting confidence, and transformed relational frustration) were summed after confirming the scores had similar distributions in our population. a Covariates in basic model: age at SRS-2, maternal education, marital status, maternal age, paternal age, child sex; covariates in full model: basic + maternal self-reported smoking during pregnancy, early exclusive breastfeeding, delivery mode, peripartum antibiotics, and gestational age b Covariates the same as a except full model adjusts for maternal urinary cotinine during pregnancy in place of maternal self-report of smoking c Covariates the same as a with the addition of an indicator variable for any breastfeeding at six months in the basic and full models for six-week microbiome analyses and in the full model for all other microbiome age analyses d Covariates the same as a with the addition of the Parenting Relationship Questionnaire (PRQ) a MaAsLin models the microbiome as the outcome. Thus, estimate is increase in % relative abundance per point increase on SRS2 T-score b Mean relative abundance of specific amplicon sequence variant given microbiome age c Covariates in basic model: age at SRS-2, maternal education, marital status, maternal age, paternal age, child sex; covariates in full model: basic + maternal self-reported smoking during pregnancy, early exclusive breastfeeding, delivery mode, peripartum antibiotics, and gestational age d Covariates the same as c except full model adjusts for maternal urinary cotinine during pregnancy in place of maternal self-report of smoking e Covariates the same as c with the addition of an indicator variable for any breastfeeding at six months in the basic and full models for six-week microbiome analyses and in the full model for all other microbiome age analyses f Covariates the same as c with the addition of the Parenting Relationship Questionnaire (PRQ)