Gut microbiome-wide association study of depressive symptoms

Depression affects 11–15% of individuals over a lifetime and its prevalence increased during the COVID-19 pandemic. Despite its burden, depression’s pathogenesis remains elusive and current treatments are suboptimal. Genome-wide association studies indicate low to moderate heritability and very small genetic effect sizes, motivating research beyond genetics to identify molecular biomarkers. Growing evidence suggests the gut microbiota influences brain activity and behavior via neural and humoral pathways, with potential translational applications in neuropsychiatric disorders. Animal studies, fecal microbiota transplantation experiments, and probiotic interventions support a role for gut microbes in mood and anxiety phenotypes. However, human studies linking gut microbiome composition to depression are few, often small, and have not consistently adjusted for key confounders. This study investigates whether gut microbiome diversity and specific taxa are associated with depressive symptom severity in a large, population-based cohort and seeks replication in an independent cohort, additionally evaluating causal relationships using Mendelian randomization.

Prior human studies have reported associations between depression and gut microbiota, including depletion of butyrate-producing genera (e.g., Coprococcus, Dialister) in depressed individuals, but many were underpowered (<60 cases) and did not adequately adjust for lifestyle and medication confounders known to affect the microbiome. A parallel multiethnic HELIUS study (NCOMMS-21-20669B) reported taxa within Christensenellaceae, Lachnospiraceae, and Ruminococcaceae associated with depressive symptoms across ethnicities after comprehensive confounder adjustment. Across studies, relatively consistent associations have been reported for genera Eggerthella, Coprococcus, Subdoligranulum, Mitsuokella, Paraprevotella, Sutterella and family Prevotellaceae, though results are often conflicting with limited overlap, underscoring the need for larger, rigorously controlled studies. Mechanistically, several implicated taxa are involved in pathways relevant to neurotransmitter metabolism (butyrate, glutamate, serotonin, GABA), supporting biological plausibility.

Design: Discovery in the Rotterdam Study (RS; N=1,054 after exclusions) and replication in the native Dutch subset of HELIUS (N=1,539). Exclusions: individuals using antidepressants and non-European subjects. Depressive symptoms were assessed using CES-D (RS; 0–60 scale) and PHQ-9 (HELIUS; 0–21 scale). Ethics approvals and informed consent were obtained. Microbiome sampling and sequencing: In RS, participants mailed stool samples in sterile tubes; samples stored at −20°C upon arrival (exclusion if transit >3 days). DNA was extracted from ~300 mg stool using an automated kit with bead beating; the V3–V4 16S rRNA regions were sequenced on Illumina MiSeq (2×300 bp). In HELIUS, participants delivered samples within 6 h or after overnight home freezing; samples stored at −20°C then at −80°C at AMC. DNA was extracted from 150 mg aliquots using repeated bead beating and purified on QIAcube; the V4 region was sequenced on MiSeq (2×250 bp) with 515F/806R primers. Bioinformatics: Raw reads from both cohorts were processed using the MiBioGen consortium pipeline: closed-reference OTU clustering (no denoising) with RDP naive Bayesian classifier (v2.12) and SILVA v128 database. Reads were rarefied to 10,000 per sample. Analyses focused on genus and higher taxonomic levels. Taxa present in <3% of samples or with read counts <0.005% of total reads were excluded. Absolute counts were log-transformed (adding 1 before transformation). Diversity metrics: Alpha diversity (species richness, Shannon index, Inverse Simpson) computed at genus level; beta diversity via Bray–Curtis distance. Beta diversity association tested using PERMANOVA. Statistical analysis: Single-taxon associations with depressive symptom scores were tested using linear regression adjusting for lifestyle (e.g., smoking, alcohol) and medical confounders (e.g., PPI, metformin); false discovery rate (FDR) <5% defined significance in RS. Significant RS findings were tested for replication in HELIUS with concordant direction, followed by meta-analysis (METAL). Random forest models (RS training, HELIUS testing) identified taxa important for predicting depressive symptoms (variable importance via % increase in MSE). Mendelian randomization (MR): Two-sample MR using MR-Base platform to test causal relationships between microbiota and major depressive disorder (MDD). For depression→microbiome, 87 depression-associated SNPs were used as instruments. For microbiome→depression, due to few genome-wide significant instruments, SNP selection threshold was relaxed to P<1e−6 to obtain >1 independent instrument per taxon. IVW was primary method; heterogeneity and horizontal pleiotropy were assessed; Steiger directionality tests were performed.

Cohorts: After exclusions, RS N=1,054; HELIUS N=1,539. Microbiome composition dominated by Firmicutes (RS 77%, HELIUS 70%), Bacteroidetes (RS 13%, HELIUS 21%), Actinobacteria (0.42% both), Proteobacteria (RS 0.48%, HELIUS 0.22%). Diversity: Alpha diversity was negatively associated with depressive symptoms in both cohorts (Shannon index: RS beta = −1.57, P = 1.5×10−7; HELIUS beta = −0.64, P = 2.84×10−2). Beta diversity associated in RS (PERMANOVA R2 = 0.003, P = 0.001) but not in HELIUS (R2 = 0.0005, P = 0.51). Taxa associations: In RS, 24 genera, 3 families, 1 class, 2 orders, and 1 phylum were associated with depressive symptoms at FDR<5%. Twelve genera replicated in HELIUS with consistent direction and showed strengthened meta-analysis signals: Sellimonas, Eggerthella, Ruminococcaceae UCG002, UCG003, UCG005, Coprococcus (Coprococcus3), Lachnoclostridium, Hungatella, Lachnospiraceae UCG001, Ruminococcus gauvreauii group, Eubacterium ventriosum group, and Subdoligranulum. Of the three associated families in RS, only family Ruminococcaceae replicated in HELIUS. Directionally, Ruminococcaceae genera were depleted with higher depressive symptoms, while several Lachnospiraceae genera were increased. Notably, Sellimonas and Eggerthella were increased with symptom severity; Coprococcus, Subdoligranulum, Eubacterium ventriosum group, and multiple Ruminococcaceae groups were decreased. Random forest: Ruminococcaceae UCG005 was the top predictor of depressive symptoms (%IncMSE highest). Other important taxa included Christensenellaceae R7 group, Lachnoclostridium, Eggerthella, Sellimonas, and Hungatella; additional predictors (e.g., Roseburia, Streptococcus, Bacteroides, Anaerotruncus, Dorea, Blautia, Veillonella, Desulfovibrio, Anaerostipes, Bifidobacterium) aligned with prior reports. Mendelian randomization: Using MDD as exposure, Eggerthella showed a significant causal effect by IVW (effect = 0.237, P = 0.027) without evidence of heterogeneity or horizontal pleiotropy; Steiger testing suggested Eggerthella is more likely causally associated with MDD. Using microbiota as exposure, Sellimonas was significant by IVW (effect = −0.046, P = 5.5×10−6) but with effect direction inconsistent with observational findings. Among 87 depression-associated SNPs, rs17641524 associated with genus Acidaminococcus after multiple testing correction; MDD genetic risk score was not significantly associated with microbiome taxa.

This large two-cohort study demonstrates that gut microbiome diversity and specific taxa are associated with depressive symptom severity in the general population, addressing a key gap left by small, confounder-limited studies. Replicated associations for 12 genera and the family Ruminococcaceae, with consistent directions across cohorts and improved meta-analysis significance, support robustness despite methodological differences in sampling and sequencing. Many implicated taxa participate in pathways relevant to neuroactive metabolites (butyrate, glutamate, serotonin, GABA), offering plausible biological links via the gut–brain axis. Depletion of butyrate producers (e.g., Coprococcus, Subdoligranulum, Ruminococcaceae groups) and enrichment of Eggerthella, Sellimonas, Lachnoclostridium, and Hungatella align with prior literature and suggest potential mechanistic contributions through neurotransmitter synthesis and immune-metabolic signaling. Random forest findings corroborated regression results and highlighted taxa important for predictive modeling of depressive symptoms. MR results point to a potential causal relationship between MDD and Eggerthella, though directionality warrants further study. Overall, the findings indicate that microbiome composition may play a key role in depression and could inform mechanistic and therapeutic exploration.

The study identifies and replicates associations between depressive symptoms and multiple gut microbial genera—particularly within Ruminococcaceae and Lachnospiraceae—and family Ruminococcaceae. It confirms prior links with Eggerthella, Coprococcus, and Subdoligranulum and highlights novel associations including Sellimonas, Lachnoclostridium, Hungatella, Lachnospiraceae UCG001, Eubacterium ventriosum group, and Ruminococcus gauvreauii group. These taxa are involved in metabolic pathways relevant to glutamate, butyrate, serotonin, and GABA. Preliminary MR suggests Eggerthella may be causally related to MDD. Future research should include longitudinal and interventional studies, standardized sampling/sequencing, integrative multi-omics (metagenomics, metabolomics), and mechanistic work to delineate causal pathways and therapeutic potential.

• Different depression rating scales used in discovery (CES-D) and replication (PHQ-9) may reduce power and comparability.
• MR analyses were underpowered due to limited genetic instruments: depression SNPs have small effects; few genome-wide significant microbiome SNPs necessitated relaxed instrument thresholds (P<1e−6), limiting causal inference strength.
• Differences in stool collection, storage, and 16S rRNA variable regions sequenced (RS V3–V4; HELIUS V4) can affect microbiome profiles and alpha diversity estimates, potentially reducing power and introducing methodological heterogeneity.
• Partial non-overlap of findings between cohorts may reflect small effect sizes and Type II error; differing taxonomic classification approaches across studies (ASVs vs closed-reference OTUs) further complicate comparisons.
• Analyses were restricted to European-ancestry participants in HELIUS replication, which may limit generalizability across ethnicities.