Human cytomegalovirus in breast milk is associated with milk composition and the infant gut microbiome and growth

Human cytomegalovirus (CMV) is a double-stranded DNA herpesvirus with high global seroprevalence among women of childbearing age and frequent reactivation in the mammary gland during lactation. Breast milk is a common source of infant CMV exposure, with high postnatal transmission rates in term infants of seropositive mothers. While postnatal CMV can cause significant morbidity in preterm or immunocompromised infants, effects in healthy, full-term infants are considered minimal. However, the relationship between CMV reactivation and human milk composition—including hundreds of nutritive and bioactive components—has been largely unexplored. The study aims to determine whether the presence of CMV DNA in human milk is associated with specific changes in milk gene expression and metabolite profiles, and whether these milk CMV-related changes are linked to the infant gut microbiome composition and early growth outcomes.

Prior work shows nearly universal CMV reactivation in the mammary gland during lactation among seropositive women and substantial postnatal transmission via milk. Clinical consequences are well documented for preterm infants, but data are sparse for healthy term infants. Limited studies have examined milk composition during CMV reactivation; one reported increased pro-inflammatory cytokines. Broader assessments of milk’s transcriptome or metabolome in relation to CMV are lacking. Early-life CMV acquisition has been associated with altered gut microbiota and increased atopy risk in some cohorts, and lower Bifidobacterium abundance is implicated in adverse outcomes. Impacts of CMV exposure/acquisition on infant growth have been mixed, with some studies in vulnerable populations linking CMV to lower length- or weight-for-age. The IDO1-mediated tryptophan-kynurenine pathway has been implicated in immune responses to CMV in other tissues.

Study design and cohort: Observational analysis within the Mothers and Infants Linked for Healthy Growth (MILK) study of full-term, exclusively breastfeeding dyads at ~1 month postpartum across two U.S. sites (MN, OK). Milk was collected by full-breast expression two hours after a complete feed, aliquoted, and frozen at −80°C. Infant fecal samples were collected at 1 and 6 months; infant anthropometrics were measured at birth, 1, and 6 months. Ethical approvals were obtained and informed consent provided. CMV detection in milk: DNA from milk cell pellets was sequenced via two pipelines: (1) human low-pass WGS (~1×), and (2) shotgun metagenomic sequencing (SMS). Reads were mapped with Bowtie2 to CMV reference genomes from human milk isolates; samples were called CMV+ if ≥1 concordantly mapped read pair with MAPQ>5 from either platform. Proportion of CMV-mapped reads was computed as a proxy for viral load. Validation was performed by qPCR (UL83 gene and IE2 exon 5) in 187 milk samples; viral load was the mean of both assays. Milk transcriptomics: Bulk RNA-seq of milk cell pellets (N=221). Differential expression between CMV+ and CMV− used DESeq2 across 17,675 genes with covariates: maternal age, pre-pregnancy BMI, self-reported race, parity, infant age (days), RNA integrity number (RIN), RNA-seq pool, and extracted RNA mass. Enrichment analysis used EnrichR (GO Biological Process). Cell-type proportion deconvolution used Bisque with a public milk single-cell RNA-seq reference to estimate immune vs luminal proportions. Milk metabolomics: Untargeted metabolomics on 1-month milk (N=142; 58 CMV+, 84 CMV−) via GC-MS, LC-MS, and LC-MS/MS. 475 metabolites identified; after filtering (>20% missing), 458 quantified. Batch correction (ComBat), log transform, median-centering, scaling. Associations with CMV status tested by multivariable linear regression including center, parity, maternal age, pre-pregnancy BMI, gestational diabetes, self-identified race, and Healthy Eating Index (averaged across timepoints). Kynurenine/tryptophan ratio derived and analyzed similarly; associations with CMV read proportion tested within CMV+ samples. Infant fecal metagenomics: Shotgun metagenomics of infant feces at 1 and 6 months (taxa and functional pathways). Abundance estimation via Shi7/BURST (GTDB r95) and HUMANN3. Data filtered (relative abundance >0.001 in ≥10% samples). Centered log-ratio transform with pseudocount for zeros. PCA performed separately for 1- and 6-month taxa and pathways. Associations between milk CMV status and metagenome PCs (explaining ≥5% variance) tested with multivariable regression including delivery mode, parity, maternal age, race, pre-pregnancy BMI, gestational diabetes, Group B Streptococcus status, fecal collection site, Healthy Eating Index, and at 6 months, exclusive breastfeeding and complementary foods. Alpha diversity (inverse Simpson) analyzed similarly. Species-level associations (56 species) tested using linear mixed-effects models (timepoint, covariates as above; random effect for dyad). Infant growth analyses: WHO Z-scores computed for weight-for-length (WLZ), length-for-age (LAZ), and weight-for-age (WAZ). Associations of CMV status with Z-scores at birth, 1, and 6 months used multivariable linear models (adjusted for the corresponding Z-score at birth [except when birth was the outcome], infant race, maternal pre-pregnancy BMI, gestational diabetes, household income, delivery mode). Within CMV+ infants, associations of CMV read proportion with Z-scores were tested. Associations of milk kynurenine with WLZ were tested with and without adjustment for CMV status, and stratified by CMV status. Structural equation modeling (lavaan) evaluated pathways linking CMV status or viral load, milk kynurenine, and infant 1-month WLZ; model fit assessed by χ², CFI, NFI, RMSEA, and SRMR.

- CMV detection in milk and validation: Of 276 milk samples, 96 (35%) were CMV+. Shotgun sequencing-based CMV calling agreed strongly with qPCR (Pearson r=0.88). Using qPCR as ground truth, shotgun detection had 92.7% sensitivity and 94.7% specificity; conversely, qPCR vs. shotgun ground truth had 87.9% sensitivity and 96.9% specificity. - Maternal characteristics: CMV+ milk was less common among mothers self-identifying as white/European-American (74% vs. 91%, P=3.1×10−4, q=3.7×10−3); other traits showed no significant differences. - Milk transcriptome: 36 genes were differentially expressed (q<0.05), 34 upregulated in CMV+ milk. Enrichment highlighted immune/antiviral pathways, especially cellular response to interferon-gamma (OR=74.5, P=5.22×10−15, q=2.70×10−12). Within CMV+ samples, CMV read proportion positively correlated with BATF2 and IDO1 expression. CMV+ milk had higher estimated immune cell proportions (mean 16.5% vs. 12.6%, P=0.041). - Milk metabolome: Kynurenine (effect=0.74, P=2.3×10−8, q=1.2×10−3) and kynurenic acid (effect=0.79, P=5.8×10−8, q=2.7×10−3) were elevated in CMV+ milk. The kynurenine/tryptophan ratio associated more strongly with CMV status (effect=0.82, P=9.4×10−7) and correlated with CMV read proportion within CMV+ samples (β=0.19, P=6.3×10−3). IDO1 expression correlated with the kynurenine/tryptophan ratio (β=0.35, P=6.9×10−8). - Infant gut microbiome: At 1 month, milk CMV status associated with taxon abundance PC3 (β=1.79, P=1.1×10−3, q=5.6×10−3), which loaded negatively on Bifidobacterium species. No associations at 6 months or with pathway PCs; alpha diversity not different. Species-level modeling across 1 and 6 months identified 9 taxa associated with CMV exposure (q<0.05): six Bifidobacterium spp. and Bacteroides fragilis were reduced, Clostridium tertium increased. Strongest association: Bifidobacterium infantis (β=−0.45, P=1.4×10−3, q=0.028). - Infant growth: Infants fed CMV+ milk had higher WLZ at 1 month (β=0.38, P=0.011; N=246), lower LAZ at 1 month (β=−0.27, P=0.025), and no WAZ difference; effects were not present at 6 months. Within CMV+ infants, higher CMV read proportion correlated with lower 1-month WLZ (β=−0.20, P=1.1×10−3) and higher LAZ (β=0.12, P=0.042); no WAZ correlation. Milk kynurenine was positively associated with 1-month WLZ (β=0.19, P=3.9×10−5), persisting after CMV adjustment (P=0.014); stratified analyses showed a positive association in CMV− (β=0.23, P=0.024) and a non-significant positive trend in CMV+ (β=0.13, P=0.31). Including both kynurenine and CMV read proportion within CMV+ infants showed opposing associations with WLZ (kynurenine β=0.22, P=0.088; CMV reads β=−0.14, P=0.013). - Structural equation modeling: Best-fit models supported a positive direct effect of milk kynurenine on 1-month WLZ and a negative direct effect of CMV viral load on 1-month WLZ within CMV+ infants, with no evidence that kynurenine mediates viral load effects or that CMV status directly affects WLZ after accounting for kynurenine.

The presence of CMV DNA in human milk was linked to discrete, biologically coherent changes in milk composition—particularly upregulation of interferon-related genes and the IDO1-driven tryptophan-to-kynurenine pathway—and to alterations in the infant gut microbiome marked by reduced Bifidobacterium species. These findings suggest an immune activation state in the lactating mammary gland during CMV reactivation, with metabolic reprogramming consistent with prior CMV studies in other tissues. The microbiome findings point to a potential ecological shift in early gut colonization among infants exposed to CMV+ milk, although functional pathway differences and alpha diversity changes were not detected, and causality cannot be inferred. Infant growth analyses revealed two opposing associations: (1) higher milk kynurenine correlated with greater 1-month WLZ (and lower LAZ), independent of CMV status, and (2) higher CMV viral load (within CMV+ milk) correlated with lower 1-month WLZ and higher LAZ. Structural equation models supported direct effects of kynurenine and viral load on WLZ but no evidence for mediation via kynurenine. These effects were modest and did not persist to 6 months, suggesting short-term influences on growth trajectories in healthy term infants. The study did not find evidence that CMV-related changes in milk composition directly explained the microbiome differences, implying distinct pathways by which CMV exposure may influence microbiota and growth. Overall, the results address the central question by demonstrating that CMV in breast milk associates with specific milk transcriptomic and metabolomic signatures, measurable shifts in the infant gut microbiome (notably lower Bifidobacterium spp.), and short-term growth differences. Given the high global prevalence of CMV, even subtle population-level effects on early microbiome development and growth merit attention.

This study shows that CMV DNA presence in human milk is associated with upregulation of interferon-responsive genes and the IDO1 kynurenine pathway, elevated kynurenine/kynurenic acid levels, decreased Bifidobacterium (including B. infantis) and Bacteroides fragilis, and increased Clostridium tertium in the infant gut at 1 month, as well as short-term differences in infant growth (higher WLZ linked to kynurenine; lower WLZ linked to higher CMV viral load within CMV+). The impacts were modest and did not persist to 6 months in healthy term infants. Future research should: (1) longitudinally track maternal and infant CMV status to disentangle exposure from transmission; (2) assess causal mechanisms linking CMV reactivation, milk immune-metabolic changes, infant microbiome assembly, and growth; (3) evaluate dose-response and timing effects of viral load; and (4) examine clinical implications in vulnerable populations (e.g., preterm infants) and potential interventions.

- Observational design limits causal inference. - Infant CMV serostatus and postnatal acquisition were not measured; congenital CMV status was unknown. - Single postpartum timepoint (~1 month) for milk composition and CMV detection; low viral load cases may have been missed by sequencing detection thresholds. - Potential residual confounding despite extensive covariate adjustment; antibiotic exposure data were incomplete. - Cohort demographics (mostly white, highly educated) may limit generalizability. - Cell-type deconvolution from bulk RNA-seq is imprecise; single-cell reference dataset CMV status was unknown.