Ecological change of the gut microbiota during pregnancy and progression to dyslipidemia

Dyslipidemia is a common metabolic disorder characterized by abnormal lipid levels and is prevalent in North America and China. Pregnancy induces metabolic and immunological changes that physiologically raise triglycerides (TG) and cholesterol (CHOL), particularly in the third trimester. Dyslipidemia during pregnancy increases risks of gestational complications and later-life cardiovascular disease. Prior work links gut microbiota richness and specific taxa to lipid levels and suggests microbiota-based interventions may influence dyslipidemia. However, whether pregnancy-related lipid changes are associated with gut microbiota dynamics, and whether mid-gestation microbiome profiles can predict dyslipidemia later in pregnancy, remain unclear. This study investigates longitudinal changes in gut microbiota during pregnancy, their associations with maternal lipid profiles and dyslipidemia, and evaluates the predictive value of mid-pregnancy gut microbiota for late-pregnancy dyslipidemia.

Design and cohort: Prospective longitudinal cohort within the Mother and Child Microbiome Cohort (MCMC). Pregnant women were recruited at the affiliated hospital of Nanjing Medical University (2017–2018). Stool samples were collected at second trimester (T2, mean 14.14 ± 0.95 weeks) and third trimester (T3, mean 32.11 ± 0.59 weeks). Negative controls (n=10) included across collection, transport, and extraction. Samples stored at −20°C then transferred to −80°C. Inclusion resulted in 513 women (median age 29.2 years). Clinical and biochemical data were extracted from the hospital information system; gestational age based on last menstrual period. Dyslipidemia was defined per AHA guidelines (CHOL ≥ 2.60 mmol/L; TG ≥ 2.26 mmol/L; LDL-C ≥ 1.41 mmol/L; HDL-C < 1.00 mmol/L) and subtyped (hypercholesterolemia, hypertriglyceridemia, hyperlipidemia when both elevated). Exclusion: pre-existing diabetes or GDM, thyroid disease, intrahepatic cholestasis of pregnancy, tumor, eclampsia, HBV; antibiotic use within 3 months before sampling; multiple pregnancy; ART; missing key clinical levels or stool samples at T2 or T3. Ethics approval obtained; written informed consent. Sequencing and profiling: 16S rRNA gene sequencing targeted V3–V4 region using primers 5′-ACTCCTACGGGAGGCAGCAG-3′ and 808R 5′-GACTACHVGGGTATCTAATCC-3′; Illumina MiSeq PE250. Quality control with QIIME2; DADA2 for truncation, filtering, denoising, merging to obtain exact ASVs (100% homology). Taxonomy assigned using SILVA v138. Decontam used to remove contaminants (131 of 39,691 ASVs). Alpha and beta diversity computed with vegan; rarefied to minimum depth 32,007 reads (range 32,007–120,183). Genera retained if relative abundance ≥0.01% across all samples and present in ≥10% of samples (98 of 457 genera) for downstream analysis. Shotgun metagenomics: DNA extracted with E.Z.N.A. Soil DNA Kit; libraries prepared (NEXTFLEX Rapid DNA-Seq); sequenced on Illumina NovaSeq 6000. KneadData v0.7.4 for trimming and filtering; human reads removed via Bowtie2 (GRCh37/hg19); QC with FastQC. Taxonomic profiling with MetaPhlAn2 v2.7.2; functional profiling with HUMAnN3 v0.11 using DIAMOND, UniRef90, and ChocoPhlAn. At T2, 459 species, 503 KEGG pathways, and 846 KOs identified before filtering; retained features after thresholds: species overall abundance ≥0.01% and present in ≥10% of samples; KOs abundance ≥1e−6; pathways abundance ≥0.01% and present in ≥10%. Final retained at T2: 176 species, 150 KOs, and 268 KEGG pathways across 141 samples (from initially 154 metagenomes, 13 low-quality removed). Network/CAG analysis: Co-abundance groups (CAGs) constructed from genera with relative abundance ≥0.10% and present in ≥10% of samples (n=457) using Ward clustering on Kendall correlations (made4 package). Networks built separately for T2 and T3; visualized in Cytoscape 3.7.1. NetShift used to identify driver taxa between T2 and T3 using neighbor shift (NESH) and betweenness centrality; correlation threshold 0.3. Statistical analyses: Paired Wilcoxon rank-sum test for within-subject lipid changes from T2 to T3. Alpha diversity via Shannon index; beta diversity via UniFrac with PERMANOVA (adonis). DMM clustering used to define community types at genus level; optimal clusters by Laplace/AIC. Between-cluster lipid differences via Kruskal–Wallis. Longitudinal associations between microbiota and lipid levels via linear mixed-effects models (lme) with subject ID as random effect; fixed effects included pre-pregnancy BMI, parity, age, and sampling time; multiple-testing correction via Benjamini–Hochberg with adjusted p<0.2 threshold; MaAsLin used for complementary association testing. Mediation analysis tested biochemical mediators between taxa and dyslipidemia using R package mediate. Data transformations: genera/species log2-transformed with pseudocount 1e−5; KOs log10-transformed with pseudocount 1e−3. Prediction modeling: Multi-class and two-level random forest classifiers (randomForest package) trained to classify dyslipidemia (hypercholesterolemia, hypertriglyceridemia, hyperlipidemia) using key lipid-associated genera, associated biochemical markers, and their combination; 500 trees; mtry = n/3; 100 bootstrap cross-validations with 80/20 (or 4:1) train/test splits. Performance measured by AUC, sensitivity, specificity, F1; CIs via bootstrap (mROC). Random forest regression models predicted T3 TG and CHOL; performance by R² and MSE. Data/code availability: Raw sequences in GSA (CRA014190). Code: https://github.com/Xia-Lab-NMU/GMD_study.

• Cohort and lipid trajectories: Among 513 pregnant women, lipid profiles changed significantly from T2 to T3 (paired Wilcoxon, p<0.001). 32.9% developed hyperlipidemia at T3; 2.6% with dyslipidemia at T2 normalized by T3.
• Microbiota dynamics: Shannon diversity increased during gestation overall; beta diversity (community composition) differed significantly between T2 and T3 (PERMANOVA p<0.001). Phylum-level shift with decreased Firmicutes and increased Bacteroidetes from T2 to T3. Dyslipidemic women exhibited significantly lower alpha diversity than healthy counterparts.
• Community structure: Five robust co-abundance groups (CAGs) identified; CAG2, CAG3, and CAG5 were enriched at T3. NetShift identified key driver taxa from T2 to T3 including Paraprevotella, Odoribacter, and Butyricimonas.
• Taxa associated with lipid profiles: Several genera were negatively associated with lipid levels and dyslipidemia risk, notably Bacteroides, Paraprevotella, Alistipes, Christensenellaceae R7 group, Clostridia UCG-014, and UCG-002. Alistipes and Christensenellaceae R7 group were significantly enriched in healthy women and remained stable across gestation.
• Species-level associations (metagenomics): TG-associated species at T2 included Parabacteroides johnsonii, Coprococcus sp. AST75.1, Clostridium baratii, Peptostreptococcaceae (unclassified), and Veillonella dispar; several remained significant at T3. CHOL-associated species at T2 included Bifidobacterium adolescentis, Bacteroides coprocola, Bacteroides faecichinchillae, and Veillonella (unclassified); consistent across T2 and T3. Additional findings: Bacteroides faecalis associated with reduced CHOL; Bacteroides coprophilus and Bacteroides fragilis positively associated with lipid levels.
• Functional pathways: Eighteen KEGG orthologs (KOs) associated with lipid levels (mainly TG). K0080 and K0185 (phosphocholine/inositol and galactyl-PI related) were primarily contributed by Faecalibacterium spp. Pathway analysis indicated depletion of IL-17 signaling and Th17 cell differentiation in hyperlipidemia, with enrichment of inositol phosphate metabolism and mineral absorption.
• Mediation by biochemical markers: Thirteen significant mediation links identified; many involved uric acid. Alistipes associated with reduced dyslipidemia risk mediated by retinol-binding protein (8.3%) and uric acid (3.7%). Bacteroides linked to reduced dyslipidemia risk mediated by uric acid (11.3%).
• Predictive modeling: Random forest classifier using key genera plus biochemical markers achieved micro-averaged AUC 0.824 (95% CI reported as 0.782–0.855) for dyslipidemia prediction, outperforming biochemical markers alone. Subtype AUCs: hypertriglyceridemia 0.796 (95% CI: 0.715–0.817), hypercholesterolemia 0.765 (95% CI: 0.607–0.878), hyperlipidemia 0.800 (95% CI: 0.714–0.865). Two-level classifier AUCs: 0.841 (hypertriglyceridemia), 0.850 (hypercholesterolemia), 0.805 (hyperlipidemia). Random forest regression predicted T3 lipids with R² = 0.613 (TG) and 0.568 (CHOL).

The study demonstrates that the gut microbiota undergoes significant ecological changes from mid- to late pregnancy and that women who develop dyslipidemia display lower alpha diversity and distinct community structures. Specific genera—particularly Alistipes, Bacteroides, Paraprevotella, Christensenellaceae R7 group, and certain UCG taxa—were consistently and negatively associated with lipid levels and dyslipidemia risk. Functional profiling implicates inflammatory and immune-related pathways (e.g., IL-17/Th17) and carbohydrate/inositol metabolism in the microbiome–lipid axis. Mediation analyses suggest that microbiota may influence dyslipidemia via biochemical intermediates, notably uric acid and retinol-binding protein, aligning with prior evidence that microbiota can modulate uric acid metabolism and systemic inflammation. Machine learning models integrating mid-gestation microbiota and biochemical markers predict late-pregnancy dyslipidemia with good discrimination, supporting the potential of microbiome-informed risk stratification. Collectively, these findings address the hypothesis that dynamic gut microbiota changes during pregnancy are linked to maternal lipid profiles and that mid-pregnancy microbiome features can help predict progression to dyslipidemia, potentially through inflammatory and metabolic pathways involving SCFA production and uric acid modulation.

This longitudinal cohort study links pregnancy-associated shifts in the gut microbiome to maternal lipid profiles and dyslipidemia, identifying stable microbial genera (e.g., Alistipes, Bacteroides, Christensenellaceae R7 group) and inflammatory functional pathways associated with lipid regulation. Mid-pregnancy microbiota combined with biochemical data can predict late-pregnancy dyslipidemia with strong performance, highlighting a potential noninvasive approach for early identification and intervention. Future research should validate these biomarkers across diverse, multi-center cohorts; elucidate causal mechanisms (e.g., SCFA-mediated inflammation and uric acid pathways); integrate immunological characterization of the microbiome; and assess microbiome-targeted preventive or therapeutic strategies during pregnancy.

• Single-center cohort limits generalizability; external validation across different populations is needed.
• Potential confounding from complex clinical and lifestyle factors; although adjusted for key covariates, residual confounding may persist.
• Functional inference focused on metagenomic KO/pathway profiling without direct immunologic characterization of host–microbe interactions.
• Species-level predictive models did not outperform genus-level models; granularity of species effects and strain-level heterogeneity may be underexplored.
• Metagenomic subset size (n≈141 at T2) smaller than full cohort may limit power for species/functional associations.