Maternal diet alters human milk oligosaccharide composition with implications for the milk metagenome

The study investigates whether maternal diet during lactation causally alters human milk oligosaccharide (HMO) composition and, in turn, the milk microbiome’s functional capacity. Human milk confers protection against neonatal and later-life diseases. Animal work suggests maternal high-fat diet during lactation can negate benefits and induce dysfunction in offspring, implying diet can modulate milk composition. HMOs, complex indigestible glycans that nourish microbes and protect the infant, vary by maternal genetics (secretor status), geography, and potentially diet. The milk microbiome, although low biomass, is diverse and may influence infant gut colonization. Prior studies associated HMO profiles with infant microbiome and growth outcomes, but a causal link between maternal diet, HMO profile, and milk microbiome function had not been demonstrated. This study tests the hypothesis that altering maternal carbohydrate type (glucose vs galactose) or energy source (carbohydrate vs high fat) rapidly changes specific HMO moieties and the functional metagenome of milk bacteria.

Previous work shows breastfeeding reduces risks of necrotizing enterocolitis and chronic diseases. Cross-fostering and maternal high-fat diet studies in rodents and primates indicate maternal diet during gestation/lactation impacts offspring microbiota and metabolism. HMOs promote beneficial bacteria, act as decoys for pathogens, and support gut barrier and development. Charbonneau et al. reported reduced total, sialylated, and fucosylated HMOs in undernourished mothers’ milk associated with stunted infant growth, reversible by sialylated HMO supplementation in gnotobiotic mice. HMO profiles vary by geography and maternal factors, and secretor genotype (FUT2) strongly shapes HMO. Infant stool microbiomes differ by feeding mode due to milk components. However, no prior study established a causal relationship between maternal diet, HMO composition, and milk microbiome function. This study builds on these findings by using controlled dietary interventions and cross-over design to isolate diet effects.

Design: Two single-blinded cross-over clinical dietary intervention cohorts of healthy lactating women (each subject serves as her own control) admitted to the General Clinical Research Center for controlled feeding and timed milk collection.

• Glu/Gal Cohort (n=7; 8–11 weeks postpartum; obese by design): randomized to receive an isocaloric, isonitrogenous liquid diet where the sole carbohydrate was either glucose or galactose (83% carbohydrate, 10% protein, 7% fat; sugar provided ~60% of estimated energy requirement). Each admission lasted 30–57 h, with a 1–2 week washout before crossover to the alternate sugar. Milk collected every 3 h; final timepoint used for analysis.
• Carb/Fat Cohort (n=7; 9–12 weeks postpartum; non-obese): randomized to isocaloric, isonitrogenous diets for 8 days: carbohydrate-rich (60% carb, 25% fat, 15% protein) versus high-fat (30% carb, 55% fat, 15% protein), with 1–2 week washout before crossover. Meals were pre-weighed/packaged; in-clinic days 5–8 with timed milk collection; day 8 final timepoint used. Sample handling: Standardized sterile collection from both breasts; pooled aliquots stored at −80°C; limited freeze-thaw cycles; strict contamination control. Secretor status: Determined by 2'-fucosyllactose (2'FL) abundance (cutoff 100 nmol/mL) to classify Secretor vs Non-Secretor; non-secretors excluded from metagenomic analyses to avoid confounding. HMO quantification: High-performance liquid chromatography (HPLC) of fluorescently 2-AB-labeled HMOs after SPE cleanup; raffinose internal standard; identification via standards and MS validation; absolute quantification via standard curves. Computed total HMO, relative abundances, and total HMO-bound fucose and sialic acid (molar sums of residues across HMOs). One high-fat subject excluded from HMO paired analysis due to insufficient volume. Metagenomics: Whole-genome shotgun (WGS) sequencing (Illumina HiSeq 2000/2500, ~2 Gb/sample). Quality control: Kneaddata for trimming and human read removal; removal of PhiX/adapters; kit negatives processed and sequenced; decontamination via prevalence-based decontam. Taxonomy: Kraken/Bracken to species-level relative abundances. Diversity: alpha (Shannon), beta (Bray-Curtis, PCoA), PERMANOVA and Mann-Whitney tests. Functional profiling: HUMAnN2 for pathway abundances; feature selection by LEfSe (LDA > 2.0, p<0.05). Marker genes: ShortBRED marker sets for fucosidase (KEGG K01206) and sialidase (K01186) against UniRef90; relative abundance normalized to mapped bacterial reads; analysis restricted to samples with >15,000 mapped bacterial reads for reliable KO quantification. Inferred metagenomics: 16S rRNA V4 sequencing (MiSeq 2x250), QIIME v1.9.1, closed-reference OTUs (Greengenes 13_5), PICRUSt to infer KO abundances (including K01206) for all samples and for cross-validation when WGS depth insufficient. In vitro assays: Growth of Streptococcus spp. (S. mitis, S. oralis, S. parasanguinis, S. cristatus, S. thermophilus) with or without individual fucosylated HMOs (2'-fucosyllactose, lacto-N-fucopentaose I at 3 mg/mL) in defined media; Staphylococcus aureus and S. epidermidis tested as controls lacking fucosidase. Growth monitored by OD600 in microplates; carrying capacity (K) computed via GrowthCurver; statistical tests (t-test or Mann-Whitney) with outlier handling. Statistics: Paired t-tests for HMO-bound residue comparisons; multiple testing correction (FDR) for individual HMOs; alpha/beta diversity via phyloseq/vegan; PERMANOVA for clustering; LEfSe for differential pathways; linear regressions for associations (reporting R and p-values); significance p<0.05.

• HMO composition shifts with maternal diet:
  • In Glu/Gal Cohort, total HMO-bound fucose was significantly lower on the glucose diet versus galactose (paired t-test p=0.03). No individual fucosylated HMO reached significance after correction, likely due to small n.
  • In Carb/Fat Cohort, total HMO-bound sialic acid was significantly lower on the high-fat diet versus carbohydrate diet (paired t-test p=0.02). Individual sialylated HMOs were not significant after FDR in paired samples.
  • Overall, some individual HMOs increased up to ~50% with galactose and ~30% with high-fat diet (e.g., FLNH), while many remained stable; non-secretors showed expected reductions in α1-2-fucosylated HMOs (e.g., 2'FL, LNFP I).
• Milk microbiome taxonomy largely stable across diets:
  • Dominant taxa included Staphylococcus and Streptococcus spp.; Bifidobacterium and Lactobacillus present at lower abundance.
  • Alpha diversity (Shannon) and beta diversity (Bray–Curtis/PCoA) did not differ by diet in either cohort (Mann-Whitney and PERMANOVA non-significant).
  • Secretor status strongly associated with taxonomic differences; non-secretors showed distinct profiles (e.g., Bacillus, Pseudomonas), thus excluded from metagenomic comparisons.
• Functional metagenome varied with diet:
  • LEfSe identified multiple metabolic pathways differing by diet in both cohorts, including amino acid biosynthetic pathways (e.g., tryptophan, histidine) and carbohydrate metabolism.
  • Heatmap analyses showed carriage of biosynthetic capacity for essential and conditionally essential amino acids among milk bacteria.
• HMO–gene function linkage:
  • Strong positive association between HMO-bound fucose concentration and bacterial fucosidase (K01206) abundance:
    • WGS ShortBRED (samples with >15,000 mapped bacterial reads): R=0.98, p=0.003.
    • Inferred metagenomics (PICRUSt) across all samples: R=0.88, p=0.0009.
  • Fucosidase abundance increased with galactose vs glucose diet (paired t-test p=0.030), mirroring increased HMO-bound fucose.
  • The diet-induced increase in fucosidase abundance was proportional to the increase in HMO-bound fucose across paired samples (R=0.88, p=0.048).
  • Species contributing fucosidase were primarily Streptococcus spp. (e.g., S. mitis, S. pneumoniae, S. oralis), consistent with WGS annotations.
• In vitro validation of functional consequences:
  • 2'-fucosyllactose (2'FL) increased carrying capacity (K) of fucosidase-positive Streptococcus abundant in milk:
    • S. mitis: p<0.0001 with 2'FL; p=0.0099 with LNFP I.
    • S. oralis: p=0.0007 with 2'FL.
    • No significant effect for S. parasanguinis (p=0.11) or low-abundance S. cristatus (p=0.28) with 2'FL.
  • No growth advantage from 2'FL for fucosidase-negative Staphylococcus spp. (S. epidermidis p=0.94; S. aureus p=0.28).

The study demonstrates that short-term maternal dietary changes during lactation rapidly and predictably modify HMO composition—specifically lowering total HMO-bound fucose with glucose vs galactose and lowering total HMO-bound sialic acid with high fat vs carbohydrate intake. Although bacterial community composition (taxonomy, alpha/beta diversity) remained stable across diets, the functional metagenome shifted, indicating that diet-driven HMO alterations shape microbial functional capacity. The tight association between HMO-bound fucose and fucosidase gene abundance, along with proportional diet-induced changes, supports a causal pathway where maternal diet alters HMO substrates that select for bacteria harboring relevant catabolic genes. In vitro data corroborate that fucosylated HMOs confer growth advantages to abundant fucosidase-bearing Streptococcus spp. but not to taxa lacking fucosidase, providing mechanistic plausibility. These findings suggest that maternal diet may tailor the metabolic potential of milk microbiota and, by extension, influence early infant microbial ecology. The detection of amino acid biosynthetic pathways in milk bacteria raises hypotheses about minor microbial contributions to milk amino acid pools or, more plausibly, seeding of infant gut communities with such capacities. Clinically, fostering Streptococcus via HMO could enhance colonization resistance against pathogens like Staphylococcus aureus, potentially reducing mastitis risk, while also benefiting infant oral/gut microbiome establishment. Overall, the data provide evidence for a diet–HMO–microbiome functional axis in human milk.

This cross-over human study shows that maternal diet during lactation is an independent, rapid modifier of HMO composition and, consequently, the functional metagenome of the milk microbiome. Glucose vs galactose and high-fat vs carbohydrate diets differentially reduce HMO-bound fucose and sialic acid, respectively, without altering community taxonomy but shifting metabolic pathways. HMO-bound fucose strongly correlates with and appears to drive fucosidase gene abundance, and fucosylated HMOs selectively enhance growth of abundant Streptococcus spp. These findings highlight a mechanism by which maternal diet can shape microbial functions delivered via milk to the infant. Future research should include larger cohorts with longer interventions, longitudinal mother–infant dyad studies to assess downstream effects on infant gut colonization and metabolism, integration of transcriptomics/proteomics to quantify gene expression and enzyme activity, and interventional trials evaluating dietary strategies to optimize milk HMOs for maternal (e.g., mastitis prevention) and infant health outcomes.

• Low-biomass milk microbiome is susceptible to environmental/kit contamination; mitigated via kit negatives, decontamination pipelines, and paired cross-over design, but residual bias cannot be entirely excluded.
• Small sample sizes and short intervention durations limit power to detect changes in individual HMOs and specific taxa; underpowered for multivariate associations among multiple HMOs and bacteria.
• WGS depth insufficient in some samples for accurate KO quantification; reliance on inferred metagenomics (PICRUSt) may introduce predictive uncertainty.
• Functional inference from DNA does not measure gene expression or enzyme activity; lack of transcriptomic/proteomic validation.
• Potential undetected subtle effects on bacterial abundances; dietary compliance at home phases not independently verified; generalizability limited by cohort characteristics (e.g., BMI differences across cohorts by design).