BGPT: Paper Review: Bidirectional relations between the maternal and infant gut microbiome and behavior

Fuel Your Discoveries

Quick Explanation Copied

Quick take — core findings & caution

Kelsey et al. present a longitudinal shotgun‑metagenomics study of 121 mother–infant dyads (514 stool samples) showing: (1) infant gut richness (Chao1) increases markedly from ~1→14 months while maternal richness remains stable; (2) multiple infant taxa and functional terms (ARGs/virulence genes) associate with infant temperament (regulation and negative emotionality); (3) maternal depressive symptoms show little consistent association with maternal taxa or functional terms; and (4) complex, often small bidirectional cross-lagged links across dyads that require cautious interpretation because of multiple comparisons, parent‑report behavioral measures, limited metabolomics, and possible unmeasured confounders (diet, antibiotic timing)

Long Explanation

Visual paper analysis — "Bidirectional relations between the maternal and infant gut microbiome and behavior"

This schematic recreates the paper's key group-level result: infants' taxonomic richness increases strongly across the first year while maternal richness shows no group‑level change — result reported by the authors using Chao1 LKT and linear mixed models

Authors describe dominance of Bifidobacterium and Escherichia early (T1–T2) shifting toward Faecalibacterium and Roseburia by 14 months, consistent with established successional patterns during weaning and solid‑food introduction

Core evidence and methodological strengths

Longitudinal design with 3 timepoints and repeated sampling per dyad (N samples = 514) increases within-subject power for temporal modelling
Shotgun metagenomics + functional annotation (GO terms, virulence, ARGs) offers higher resolution than 16S for functionally relevant signals (ARGs/virulence) and strain-level inference potential

Key limitations, blindspots & sources of potential bias

Behavioral outcomes are parent‑report instruments (IBQ‑R/ECBQ, EPDS). Parental mood biases reporting and can inflate associations between maternal mood and perceived infant temperament; authors note this and call for observational measures
Multiple testing risk: Maaslin2 biomarker discovery across many taxa/functional terms produces many comparisons; single unexpected between-dyad hits (Romboutsia timonensis ↔ maternal depression) required appropriate skepticism as authors state
Important confounders incompletely resolved in main models: infant diet (breast/formula/solids), antibiotic exposure timing (labor vs postnatal), and freezing delay (samples frozen within 24 h) — all can alter microbial profiles and were handled via PCA covariates rather than explicit causal mediators
Missing metabolomic data — limits mechanistic links between taxa/ARGs and CNS‑relevant metabolites (SCFAs, Trp metabolites, bile acids) that mediate gut→brain effects; authors propose metabolomics for next steps

Mechanistic plausibility: the literature supports multiple plausible routes whereby infant microbiota could influence behavior (immune signaling, tryptophan/AhR ligands, SCFAs, vagal signaling), and maternal microbiome transfers (milk, skin, vaginal) seed infant communities — but causality requires interventional or metabolomic mediation evidence

How to improve / next experiments to test causality

Integrate longitudinal stool metabolomics (targeted SCFAs, Trp‑pathway metabolites, bile acids) taken at same timepoints to test mediation (taxa→metabolite→behavior).
Collect objective infant behavioral/physiological measures (lab observation of affect/regulation, heart-rate variability, stress reactivity) alongside parent reports to reduce reporter bias.
Use higher‑resolution modelling of antibiotic timing and feeding (daily logs) and treat them as time-varying covariates or explicit mediators in cross‑lagged models.
Attempt replication in a larger, more diverse cohort and/or apply causal inference techniques (instrumental variables, sibling designs) and, where ethical, randomized interventions (e.g., targeted prebiotic supplementation with blinded followup) with metabolomic endpoints.

Key source documents used (select)

If you want I can: (A) build a reproducible analysis notebook that re-runs the authors' Maaslin2 and BRI‑CLPM steps on their deposited NCBI reads and metadata (requires data access request steps they described), (B) add targeted metabolomic mediation models (if you provide metabolomics), or (C) run sensitivity analyses stratifying by antibiotic timing and feeding modality — select one and press "Run AI Science Analysis" to start an automated agent that will run the pipelines and produce figures/tables.

Feedback:

Updated: January 26, 2026