Changes of gut microbiota reflect the severity of major depressive disorder: a cross-sectional study

Major depressive disorder (MDD) is prevalent, recurrent, and costly. Treatment decisions depend on baseline severity; mild to moderate cases may be managed conservatively, whereas severe cases often require intensive interventions. Misdiagnosis can lead to ineffective treatment and heightened risk, including suicidality, underscoring the need for objective biomarkers to stratify severity. Emerging evidence implicates the gut microbiota in mental disorders via the gut–brain axis. Prior microbiome studies using 16S rRNA sequencing have reported disorder-specific taxa shifts (e.g., Bacteroides in unipolar depression, Prevotella in bipolar disorder) and distinct profiles between active and remitted MDD, but findings are inconsistent across studies and limited to genus-level resolution. To address these gaps, the authors applied shotgun metagenomic sequencing to characterize species-level compositional and functional changes in treatment-naive MDD across severity (mild, moderate, severe) versus healthy controls, test whether overall signatures differ by severity, integrate taxa-function networks related to severity, and identify microbial markers for severity discrimination.

The paper reviews evidence linking gut microbiota to psychiatric conditions, highlighting microbiota’s role in brain function and behavior through the gut–brain axis. Prior 16S rRNA studies in mood disorders found that Bacteroides is a perturbed hub in unipolar depression, whereas Prevotella characterizes bipolar depression. Other studies observed increased Alistipes and Anaerostipes and depletion of Dialister in current active MDD, with Bilophila enriched in milder symptoms; yet another reported increases in Bacteroidetes, Proteobacteria, Actinobacteria and reduced Firmicutes in both active and remitted MDD. These mixed results suggest microbiota differences may relate to symptom severity, but 16S approaches limit species-level resolution and functional inference. The authors position metagenomics as necessary to refine taxonomic resolution, quantify functional pathways (KEGG KOs), and improve biomarker discovery for MDD severity stratification.

Study design and participants: Cross-sectional analysis of stool metagenomes from 138 treatment-naive MDD patients and 155 healthy controls (HCs) from a prior clinical cohort. Diagnosis per DSM-IV. Severity classified by HAMD-17: mild (8–16), moderate (17–23), severe (≥24). Demographics (age, BMI, sex) were similar across HCs and MDD subgroups. Exclusions: bipolar/schizophrenia/other Axis I disorders; serious chronic somatic diseases (e.g., diabetes, cardiovascular, thyroid disease, cancer); alcohol/substance abuse or acute intoxication; pregnancy/breastfeeding; diet change or antibiotics within 1 month before sampling. Ethics approval obtained; informed consent secured. Sample collection and DNA extraction: Fresh morning stool (7–10 am) collected into sterile tubes, held at 4°C and transferred to −80°C within 6 hours. DNA extracted using E.Z.N.A. Soil DNA Kit. DNA quantity/quality assessed via TBS-380, NanoDrop2000, and 1% agarose gel. DNA fragmented to ~300 bp (Covaris M220). Libraries prepared with NEXTFLEX Rapid DNA-Seq and sequenced paired-end on Illumina NovaSeq. All samples sequenced in a single batch to avoid batch effects. Bioinformatics and quality control: Low-quality reads (length <50 bp, homopolymers >10 bp, or ambiguous bases) filtered with Sickle. Reads aligned to human genome (BWA); host sequences removed. Clean reads assembled into contigs with MEGAHIT (retain contigs ≥300 bp). ORFs predicted with MetaGene. Non-redundant gene catalog generated by clustering predicted genes at 95% identity using CD-HIT. Reads mapped back to representative sequences using SOAPaligner; gene abundance quantified as RPKM. Taxonomic and functional annotation: Gene catalog annotated against NCBI database using DIAMOND (v0.8.35); taxonomy assigned to the highest-scoring hit. Functional annotation against KEGG (KOBAS/KEGG; E-value ≤1e−5); KO abundances computed as the sum of constituent gene abundances. Community analyses: Alpha diversity (Dominance, Simpson, Shannon, Evenness) computed with PAST 4.0. Beta diversity assessed via Bray–Curtis distances at species level; PCoA performed and group differences tested by PERMANOVA. Enterotypes inferred at genus level using Dirichlet multinomial mixtures (DMM), with optimal cluster number by Calinski–Harabasz index. Differential features and biomarkers: High-abundance species prefiltered (prevalence >20%, mean relative abundance >0.01%); unclassified species excluded. LEfSe (LDA >2.5; p<0.05) identified differentially enriched taxa and KOs between HCs and each MDD severity subgroup. Random Forest classifiers with 5-fold cross-validation trained on differentially enriched species to build severity-discriminating panels; performance evaluated by ROC/AUC. Feature importance by Gini index (>0.02) used to select a 37-species biomarker panel. Co-occurrence networks and taxa–function correlations: SparCC computed correlations among differentially enriched species and KOs (p<0.05; r²>0.25). Networks visualized in Cytoscape; correlation heatmaps constructed for taxa–KO pairs with strong associations. Statistics: SPSS v22.0. Continuous variables: one-way ANOVA (mean±SD) with LSD post hoc or Kruskal–Wallis (mean±SEM) with Holm–Bonferroni correction for pairwise comparisons. Categorical variables: chi-square. Significance threshold p<0.05.

- Cohort: 138 untreated MDD (mild n=24; moderate n=72; severe n=42) and 155 HCs; no significant differences in sex (p=0.71), age (p=0.62), or BMI (p=0.71) across groups. - Alpha diversity: Simpson index significantly decreased in moderate and severe MDD; Shannon and Evenness reduced in moderate MDD. Mild did not differ significantly from HCs, indicating diversity loss associates with greater severity. - Beta diversity: PCoA showed significant separation of moderate vs HCs and severe vs HCs (PERMANOVA p=0.001 for both). Mild MDD did not separate from HCs; MDD subgroups were not significantly different from each other. - Taxonomic shifts (genus level): Bacteroides enriched in moderate and severe MDD. Faecalibacterium and Escherichia decreased in moderate MDD. Ruminococcus and Eubacterium decreased predominantly in severe MDD. - Enterotypes: Five enterotypes identified, including classical Bacteroides and Prevotella, plus Blautia- and Faecalibacterium-dominated types. HCs were enriched for Faecalibacterium enterotype (31.6%). MDD subgroups were enriched for two Bacteroides enterotypes (mild 33.3%, moderate 36.1%, severe 31.0%). - Differential species (LEfSe): Compared to HCs, mild showed 14 differentially enriched species, moderate 60, and severe 74. Across all contrasts, 99 species were differentially enriched. Unique species counts: mild 3, moderate 18 (including 4 Bacteroides), severe 32 (including 5 Bacteroides, 4 Clostridium, 6 Ruminococcus). Bacteroides (16 species moderate; 18 species severe) were consistently increased; Blautia, Clostridium, Eubacterium, and in severe, Ruminococcus were decreased. - Functional shifts (KOs): Moderate MDD had 3 differential KOs; severe had 5. Two KOs were unique to severe: K01190 (lactase) and K12373 (hexosaminidase). K12373 and K21572 (susD) positively correlated with many Bacteroides species and negatively with Blautia and Ruminococcus. - Co-occurrence networks: Bacteroides showed strong negative correlations with depleted genera (Blautia, Ruminococcus, Eubacterium), suggesting antagonistic relationships. Severe MDD exhibited pronounced depletion clusters dominated by Ruminococcus, Eubacterium, Blautia, and Clostridium with complex positive intra-cluster correlations and strong negative associations with Bacteroides. - Biomarker performance: A 37-species Random Forest panel discriminated severity pairs with excellent accuracy (AUC 0.992–0.998): mild vs moderate AUC 0.992; mild vs severe AUC 0.998; moderate vs severe AUC 0.992. - Sex-specific analyses: Both females and males with MDD showed reduced diversity and separation from HCs; compositional trends (increased Bacteroides; decreased Ruminococcaceae/Faecalibacterium) were similar. Males exhibited more differentially enriched species, many within Bacteroides.

This study addressed whether gut microbiota dysbiosis associates with MDD severity. Species-level metagenomics revealed that microbiome alterations intensify with symptom severity: diversity loss and pronounced shifts in community structure were present in moderate and severe MDD but not in mild cases. Enrichment of Bacteroides and depletion of SCFA-producing and putatively beneficial taxa (Blautia, Eubacterium, Ruminococcus) characterized more severe states. Network analyses indicated antagonistic relationships between Bacteroides and depleted commensals, suggesting community reorganization from a balanced to dysbiotic state as severity increases. Functional (KO) correlations linked Bacteroides to carbohydrate utilization (susD) and hexosaminidase pathways, implying altered glycan metabolism may accompany taxonomic shifts along the severity spectrum. These findings support the concept that gut microbial ecosystems and functions change with clinical severity and may contribute to pathophysiology via immune and metabolic routes. The robust 37-species classifier demonstrates the feasibility of non-invasive microbiota-based tools for stratifying MDD severity, potentially informing personalized interventions and monitoring. Sex-stratified analyses suggest broadly similar dysbiosis patterns across genders, with possible quantitative differences warranting larger studies.

The study delineates unique and shared gut microbiota perturbations across mild, moderate, and severe MDD, showing that dysbiosis—marked by Bacteroides enrichment and depletion of SCFA-producing commensals—tracks with increasing severity. Co-occurrence networks reveal potential antagonism between Bacteroides and beneficial genera, and functional correlations implicate carbohydrate metabolism pathways. A 37-species microbial panel accurately discriminates severity subgroups (AUC 0.992–0.998), highlighting a promising non-invasive approach for objective severity assessment and stratification. Future research should validate these biomarkers longitudinally, test their responsiveness to treatment, explore mechanistic links (including in animal models), and assess probiotic or microbiota-targeted interventions as adjuncts, especially in severe MDD.

- Sample size for severity subgroups was relatively small, and all samples were from a single clinical center, so regional variation cannot be excluded. - All patients were medication-naive; it remains unknown whether the biomarker panel can monitor treatment response; longitudinal validation is needed. - Given evidence that fecal microbiota transplantation can transfer depressive phenotypes, mechanistic studies in animal models are needed to link specific microbiota disturbances to severity and elucidate underlying pathways.