Pumping and hygiene practices are associated with bacterial load and microbial composition in human milk expressed at home

Abstract

Background

Human milk (HM) harbors a unique microbiome that contributes to the development of the infant gastrointestinal microbiome and influences long-term health outcomes. While pumping and bottle-feeding HM are increasingly common, doing so may introduce exogenous bacteria, altering this microbial community. The extent to which real-world pumping and hygiene practices alter the HM microbial community remains inadequately characterized.

Methods

We conducted a secondary analysis of 104 paired milk samples from 52 healthy women to investigate the associations between at-home pump hygiene practices and the HM microbiota. We compared samples expressed with personal equipment, allowing women to follow their typical practices, noting variations in breast pump types (closed- vs. open-systems), pre-pumping handwashing, and collection kit cleaning practices (OWN). Milk was also expressed with hospital-grade pump with new, commercially sterilized equipment kits under study-controlled conditions (STER) to serve as a control representing each woman’s own unaltered milk microbiota. Microbiota composition was characterized using aerobic culture and 16S rRNA gene sequencing.

Results

Among OWN milk samples, personal breast pump type had little impact on the HM microbiota. Pre-pumping handwashing, practiced by only 22% of participants, was associated with lower bacterial counts. Compared to samples expressed with handwashed kits, those expressed using home-sterilized kits yielded fewer total and gram-negative bacterial counts, and lower relative abundances of Proteobacteria. The microbiota of OWN milk samples expressed with home-sterilized kits more closely resembled STER samples even in the absence of pre-pumping handwashing (R² = 0.36; P < 0.001).

Conclusions

At-home hygiene practices, particularly collection kit cleaning methods, substantially influence the HM microbiota. Home sterilization of collection kits may minimize changes to the HM microbiota during expression. These findings support evidence-based recommendations for hygienic pumping practices and underscore the need for further research on the health implications (if any) of pumping-dependent variations in the HM microbiota on infant health.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12967-025-06967-5.

Introduction

Human milk (HM) harbors a unique microbiome thought to play a role in shaping infant health, contributing to the development of oral and gastrointestinal microbiomes [1–3], offering protection against opportunistic pathogens, and improving intestinal barrier function [2, 4–8]. These multifaceted roles may contribute to the benefits of breastfeeding compared to formula-feeding, including reduced incidences of infections, diarrhea, sepsis, and necrotizing enterocolitis (NEC) [5, 9–11].

While the physiological roles of the HM microbiome is an active area of study, the increasing prevalence of pumped-milk feeding practices [12–18], such as pumping and bottle-feeding HM, raises important questions about their impact on the milk microbiome and (if any) subsequent infant health outcomes. Recent studies have highlighted the association between pumped-milk feeding and a reduction in the number of bacterial species shared between a mother's milk and her infant's stool [19]. There may be a biological explanation for this as storage conditions can influence the microbial composition of pumped milk [20]. Nevertheless, the impacts of various pumping and hygiene practices on the HM microbial community remain inadequately characterized, representing a critical knowledge gap in the field.

To begin to fill this gap, we conducted the Milk in Life Conditions (MiLC) trial, a real-world evidence, in-home randomized, crossover trial comparing the microbiota in milk expressed with participants’ personal supplies (OWN) versus a hospital grade pump and new, sterile collection kits (STER) [21]. Our earlier findings revealed that milk collected with OWN supplies often had higher bacterial counts and taxa not observed in paired milk samples collected using STER supplies [21, 22]. However, not all samples obtained with OWN supplies yielded an altered microbial community, suggesting that specific pumping conditions may have a more significant impact on the HM microbiota than others.

In this secondary analysis, we investigated associations between specific at-home pumping and hygiene practices and the HM microbiota. We leveraged aerobic culture and 16S rRNA sequencing data previously reported alongside observational and survey data not previously reported. We analyzed associations between OWN sample microbiomes and hygiene practices, namely pre-pumping handwashing, and collection kit cleaning method. Each woman’s STER milk was used as a reference representing their own unaltered milk microbiota and served as a proxy for the microbial exposures infants receive when nursing directly at the breast.

Materials and methods

Study design and participants

This secondary analysis used de-identified data from MiLC, an in-home, randomized, crossover trial (https://clinicaltrials.gov/ct2/show/NCT03123874) conducted between June and October 2017 [21]. Only healthy mother and term infant dyads were included. Written informed consent was obtained according to the study protocol and approved by the Institutional Review Board at Cornell University (1608006566).

The trial was intentionally designed to capture real-world evidence about how at-home practices for human milk expression, handling, and storage influences its microbial community. To preserve the real-world relevance of our findings, no educational intervention or instruction was given prior to milk collection. Participants used their own breast pumps and hygiene practices as they typically would in their daily routines. This decision also minimized reactivity bias, also known as the Hawthorne effect, in which participants alter their behavior due to awareness of being observed or studied [23]. By observing natural behavior without prompting, we aimed to generate insights that reflect actual maternal hygiene practices and the resulting microbial exposures experienced by infants fed pumped human milk at home.

A glossary of abbreviations is located in Table 1.

Table 1.

Glossary of abbreviations

Abbreviations	Definitions
HM	Human milk
OWN	Own supplies (participant's personal pump and collection kits)
STER	Sterile supplies (hospital-grade pump and sterile collection kits)
WA	Washed (kit cleaning method)
HS	Home-sterilized (kit cleaning method)
CS	Commercially-sterilized (kit cleaning method or control condition)
UWA	Unwashed hands and hand-washed kits (composite hygiene variable)
UHS	Unwashed and home-sterilized (composite hygiene variable)
CLC	Complete linkage clustering
DMM	Dirichlet multinomial mixtures
PCA	Principal components analysis
PERMANOVA	Permutational multivariate analysis of variance
ASV	Amplicon sequence variant
HMBANA	Human Milk Banking Association of North America
LPS	Lipopolysaccharide
NEC	Necrotizing enterocolitis
PCR	Polymerase chain reaction
CDC	Centers for Disease Control and Prevention

Milk collection and microbiota analyses

Milk collection protocols and microbiota analyses have been described in detail previously [21]. Briefly, each of the 52 MiLC participants pumped twice on the same day, once with their OWN supplies and once with STER supplies (defined above). The order of collection was randomized to reduce order effects. All milk samples (OWN and STER) were collected in participants’ homes. To control for diurnal variation in milk composition and microbial load, both samples were collected during the morning hours, within a standardized 3.5-h window between 07:00 and 11:00. After the first sample was expressed, it was immediately transported to the lab and processed within 2 h of expression, in accordance with our protocol [21]. The study team then returned to the participant’s home to collect the second sample. This design ensured within-subject comparison of hygienic conditions while maintaining strict timing and handling protocols.

Milk was cultured aerobically for total aerobic (plate count agar, Becton Dickinson, Franklin Lakes, NJ) and gram-negative bacteria (MacConkey agar, Becton Dickinson, Franklin Lakes, NJ). Milk was cultured on plate count agar and MacConkey agar (both from Becton Dickinson). Serial tenfold dilutions were prepared in 0.1% sterile peptone water (Hardy Diagnostics); 10-μL drops were plated in triplicate using the drop plate method [24]. Undiluted milk (100 μL) was also spread-plated onto 100-mm plates of each agar type. All plates were incubated aerobically at 37 °C for 48 h. Colony-forming units (CFU/mL) were calculated from plates yielding ≥ 1 colony. Samples were considered culture-negative if no growth was observed on both drop and spread plates.

After DNA was extracted, the V1-V3 region of the 16S rRNA gene was amplified via PCR and sequenced on a MiSeq (Illumina Inc., San Diego, CA). Amplicon sequences were preprocessed in Quantitative Insights Into Microbial Ecology 2 (QIIME2) (v. 2018.2) [25] to produce an amplicon sequence variant (ASV) table as previously described [21]. The previously published ASV table was then reanalyzed in the context of pumping and hygiene practices.

Variables for pumping and hygiene practices

During collection of OWN milk, the study director (SMR) recorded information about personal breast pump type, pre-pumping handwashing practices, and milk collection kit cleaning practices, and as described briefly here:

• Personal breast pump type: Open- vs. closed-system type was determined from the manufacturers’ product description. Closed-system pumps have a barrier between the pump mechanism and the collection tubing, whereas open-system pumps do not.

• Pre-pumping handwashing were directly observed and recorded by the study director, categorized as washed vs. unwashed hands based on whether participants washed their hands prior to handling pump parts and expressing milk. No prior instruction was given nor was knowledge of CDC guidelines for pumping and handling milk assessed; participants assembled and used their own pumps as they typically would at home. In contrast, STER milk was collected from the same participants under controlled conditions: the study director wore gloves sanitized with alcohol-based hand sanitizer and passed new, commercially sterile collection kits to the participant, who only touched the outside of the bottle. This ensured hand microbes were not introduced onto the internal pump or bottle surfaces. This STER milk served as a comparison to OWN milk.

• Kit cleaning practices for OWN supplies were self-reported and categorized as rinsed (cleaned with water only), washed (cleaned by hand with water and detergent), or home-sterilized (HS: washed in the dishwasher, boiled, or steam-sterilized using a microwave sterilization bag). Specific dishwasher settings were not recorded. Given the limited sample size and our goal to capture real-world practices, we included all methods of sterilization into one category. Total aerobic bacteria (colony-forming units [CFU]/mL) in milk did not differ in OWN milk collected from washed (n = 32) and rinsed collection supplies (n = 7). Thus, both were collapsed into a single group [“washed” (WA)] (Tables S1 and S2). Collection supplies used for STER milk served as a reference and were classified as “commercially-sterilized” (CS).

To better reflect the multifaceted nature of at-home milk collection, we created composite variables representing the most commonly co-occurring hand hygiene and kit cleaning practices. This approach was based on observed patterns in our study population and was designed to capture real-world combinations of exposure. Specifically, we defined two composite groups:

1. Unwashed hands, paired with handwashed collection kits (UWA)

2. Unwashed hands, paired with home-sterilized collection kits (UHS)

These groupings allowed us to examine how typical combinations of practices, rather than isolated behaviors, influence bacterial load and milk microbiota composition. Other practice combinations were too infrequent to support reliable statistical comparison and were therefore excluded from composite group analyses.

Outcomes

The primary outcomes were total aerobic and gram-negative bacterial counts (CFU/mL, quantified using aerobic culturing), 16S-based bacterial alpha (α)-diversity measured as Faith’s phylogenetic diversity [26] and the relative abundances of bacterial taxa.

We also report a dichotomous variable of total aerobic bacterial counts > 10⁴ CFU/mL (yes/no), and beta (ß)-diversity using principal components analysis (PCA), complete linkage clustering (CLC) [27], and Dirichlet multinomial mixtures (DMM) [28].

Statistical analyses

Associations among categorial variables (e.g., personal pump breast pump types, pre-pumping handwashing practices, and kit cleaning practices) were tested using χ² tests [29]. Welch’s t-tests were used to examine associations between log-transformed bacterial counts and these categorical variables. Bacterial counts were transformed using log(x + 1) CFU/mL to accommodate zero counts [30]. Potential interactions were evaluated using stratified analyses. Binary outcomes of bacterial counts, namely milk yielding > 10⁴ of total CFU/mL or being culture-positive for gram-negative bacteria, were tested using χ² or logistic mixed effects regression, adjusted for study design factors (infant diet and randomized pumping order with participant as a random effect) [21]. DESeq2 was used to identify differentially abundant bacterial taxa in 16S analyses [31].

To characterize the phylogenetic profiles of bacteria in milk expressed using the two most commonly co-occurring pumping and hygiene practice combinations, we measured α-diversity using Faith’s phylogenetic diversity index [26] and compared groups using linear mixed-effects models, adjusted for study design factors. ß-diversity was evaluated via PCA of the ASV table aggregated at the genus level in Clustvis [32]. Group differences were tested by permutational analysis of variance (PERMANOVA) on the first principal component as determined via PCA and reported as R² with the associated overall P value. To ensure that ß-diversity observations were not dependent on a single algorithm or index, community structure was also assessed using CLC [27] and DMM [28] on the ASV table, which independently groups samples based on community membership without a priori assumptions about sample groupings based on pumping practices. CLC and DMM groupings were visualized with a dendrogram and a heatmap produced using the heatmap.plus package in R [33]. We compared bacterial counts across CLC and DMM groups using linear mixed effects models adjusted and corrected as described above.

We used complete case analyses for all statistical tests, excluding participants only from specific analyses where relevant data were missing. Analyses were performed between April 1, 2018 and September 22, 2020, using R (v. 3.4.4 and 3.5.3) [34]. Unless otherwise noted, values reported are means ± standard deviation (SD). Linear mixed-effects model results were considered significant at P < 0.05; all other results were considered significant at P < 0.05 after Benjamini–Hochberg false discovery rate (FDR) correction [35].

Results

Participant characteristics and frequency of pumping and hygiene practices

We analyzed 104 paired milk samples collected from 52 participants. Mothers were on average 34 ± 4.0 years old and 6.2 months postpartum (IQR: 3.3, 8.7) at the time of sampling (Table 2). Most (86%) self-identified as Caucasian. Median maternal BMI was 25.4 kg/m² (IQR:, 22.4, 28.4), and median gestational weight gain was 30.0 pounds (IQR: 25.0, 36.2). Thirty-five percent of mothers were primiparous.

Table 2.

Characteristics of participants overall and by randomized assignment

Characteristics	All Participants1 (n = 52)	OWN SUPP
		Home-sterilized (n = 13)	Washed (n = 39)
Mothers’ age in years; Mean ± SD	34 ± 4.0	33 ± 4.3	34 ± 3.9
Time postpartum in months; Median (IQR)	5.7 (3.4, 9.6)	5.7 (2.7, 8.2)	5.7 (3.5, 10.2)
Maternal race, Caucasian n (%)	45 (86)	13 (100)	32 (82)
Parity, Primiparous n (%)	18 (35)	4 (31)	14 (36)
Gestational weight gain in pounds (GWG); Median (IQR)	30.0 (25.0, 36.2)	30.0 (25.0, 30.0)	30.0 (27.0, 40.0)
Maternal Postpartum BMI; Median (IQR)	25.4 (22.4, 28.4)	25.6 (24.8, 27.4)	24.6 (21.8, 28.6)
Infant exclusively fed HM n (%)	27 (52)	9 (69)	18 (46)
Pre-pumping Hygiene Practices
Washed hands before pumping, yes n (%)	8 (15)	2 (15)2	6 (17)
	–
Personal pump, open-system	32 (62)	10 (77)	22 (56)
Infant fed HM only	27 (52)	9 (69)	18 (46)

Values reported as Mean ± SD for continuous variables or n (%) for categorical variables. HM, human milk; OWN SUPP, own pumping supplies

¹All participants used commercially sterilized (STER SUPP)

²n = 10

Open-system personal electric breast pumps were owned by 62% of participants (Table 2, Table S3). Most mothers reused a pump from a previous child (58%), while others had purchased a new pump for their current child (27%), or used a secondhand pump (6%). Eight percent were unsure if the pump they were using was new, reused, or secondhand, often because they owned multiple pumps and could not recall which was used during sampling. Only one participant used a rented hospital-grade pump, and one used a wearable pump.

Overall, 25% of participants sterilized their milk collection kits at home. The proportion of mothers who sterilized their kits was higher among open-system than closed-system pump owners, though this difference was not statistically significant (15 vs 31%, respectively, P = 0.32, χ² test).

Pre-pumping handwashing was recorded for 49 out of the 52 participants during collection of OWN milk. For three participants, hand hygiene observations were not recorded due to recording omission during the home visit. Among the 49 participants with observed data, only 11 (22%) washed their hands before pumping. Handwashing was 3.7-times more common among participants who used open- versus closed-system pumps (22% vs. 5.9%, respectively; P = 0.02, χ² test).

Over half of infants were female (52%), including two sets of twins. Nineteen infants were exclusively fed human milk, with no exposure to complementary foods at the time of sampling.

Personal breast pump type and microbiota of pumped milk

Overall, personal breast pump type (open- vs. closed-system) was not independently associated with the number of culturable bacteria or percent abundance of bacterial taxa (16S sequences) in pumped milk (Fig. 1). While milk expressed with open-system pumps tended to have higher bacterial counts, these differences were not statistically significant at a P < 0.05 threshold. For total aerobic bacteria, the mean difference was − 0.57 log CFU/mL (95% CI − 1.19 to 0.05, P = 0.08). For gram-negative bacteria, the mean difference was − 0.43 log CFU/mL (95% CI − 1.76 to 0.89; P = 0.52). Similarly, pump type was not significantly associated with the percent of milk samples that were culture-positive for gram-negative bacteria (55% vs 47% for closed- vs. open-system pumps, respectively; P = 0.55, χ² test).

Fig. 1.

Associations among maternal hygiene practices and bacterial composition of expressed human milk (n = 104). Circle size represents the mean total aerobic bacterial load (log₂ CFU/mL), and circle color indicates mean relative abundances of dominant phyla identified via 16S rRNA gene sequencing. Orange boxes highlight hygiene comparisons where both CFU counts and microbial composition (16S sequences) differed significantly (P < 0.05); yellow boxes indicate a significant difference in either CFU or 16S data alone. Results are summarized in Table S4. Combined hygiene practices (e.g., UWA, UHS) are not visualized here. *Handwashing data were missing for three participants who used closed-system pumps. Both cleaning and handwashing behaviors were independently associated with variation in milk bacterial load and microbial community composition, underscoring the influence of modifiable maternal hygiene practices

Pre-pumping handwashing practices and the microbiota of pumped milk

Pre-pumping handwashing was not significantly associated with aerobic bacteria counts in OWN milk. Mean total aerobic counts were higher in samples collected without pre-pumping handwashing [mean (SD): 1.7 × 10⁵ CFU/mL (6.5 × 10⁵)] compared to those collected after handwashing [5.8 × 10³ (1.1 × 10⁴)]; however this difference was not statistically significant (P = 0.12). The estimated mean difference was 1.6 × 10⁵ log CFU/mL (95% CI − 6.1 × 10⁴ to 3.9 × 10⁵ CFU/mL). Notably, however, the average bacterial counts of OWN milk collected without pre-pumping handwashing was higher than 10⁴ CFU/mL, the Human Milk Bank Association of North American’s (HMBANA) threshold for “contamination” [36].

Kit cleaning practices and microbiota of pumped milk

Milk samples expressed with hand-washed (WA) collection kits yielded significantly higher concentrations of cultural aerobic bacteria compared to those expressed with either home-sterilized (HS) or commercially-sterilized (CS) collection kits (P < 0.0001, Fig. 2). This difference was largely attributed to the higher proportion of WA samples (51%) exceeding the > 10⁴ total CFU/mL threshold compared to only 15% of CS or HS samples. Compared to milk collected with CS collection kits, milk collected with WA collection kits had 14 times higher odds of exceeding > 10⁴ total CFU/mL [WA vs CS, adjusted OR: 14; 95% CI 2.3, 83, P = 0.004].

Fig. 2.

Handwashed milk collection kits were associated with higher bacterial loads in milk than home- or commercially-sterilized kits. Boxplots display distributions of total aerobic and gram-negative CFU/mL among milk samples collected with three cleaning methods: handwashed (WA), home-sterilized (HS), and commercially sterilized (CS). Bold lines indicate medians; boxes span the interquartile range; whiskers reflect the full range; and dots represent outliers. Differences between groups were assessed using Welch’s t-tests. Significant pairwise differences are denoted by different letters (P < 0.0001). Although handwashing of milk collection kits was effective for some individuals, it was less consistently associated with lower bacterial counts than sterilization. On average, milk collected with handwashed kits had higher levels of culturable bacteria compared to milk collected with home- or commercially sterilized kits, suggesting greater microbial transfer during the pumping process

Comparisons using Welch’s t-tests confirmed these findings: total aerobic counts were on average 2.1 log CFU/mL higher in milk from WA kits versus CS kits (95% CI 1.1–3.0, P < 0.0001), and 2.0 log CFU/mL higher compared to HS kits (95% CI 0.4–3.5; P = 0.0007). No significant difference in total aerobic counts was observed between milk expressed with HS or CS kits (mean difference: 0.1 log CFU/mL; 95% CI − 1.1 to 1.4; P = 0.84).

Results for gram-negative bacteria showed similar patterns (Fig. 2). Milk from WA kits had 5.6 log CFU/mL higher counts than milk from CS kits (95% CI 4.0–7.1; P < 0.0001), and 4.8 log CFU/mL higher than HS samples (95% CI 1.7–8.0; P = 0.0002).

16S analysis further corroborated the influence of kit cleaning practices on HM microbial composition. Samples expressed with WA kits yielded a substantially higher relative abundance of Proteobacteria genera, including Acinetobacter, Pseudomonas, Stenotrophomonas, and Enterobacter, compared to either CS or HS samples (P < 0.05, Table S4). Specifically, Acinetobacter abundance was 4.5-fold higher in WA versus HS samples (mean difference: 13.1%; 95% CI 3.2%-22.9%), Pseudomonas was 3.3-fold higher (mean difference: 2.3%, 95% CI 0.03%-4.6%), Stenotrophomonas was 79-fold higher (mean difference: 7.0%; 95% CI 1.8%-12.2%), and Enterobacter was tenfold higher (mean difference: 1.3%; 95% CI 0.1%-2.4%). In contrast, HS samples exhibited only modest elevations in these genera relative to CS samples: Acinetobacter (3.73% vs 0.16%), Pseudomonas (0.99% vs. 0.24%), Stenotrophomonas (0.09% vs 0.04%), and Enterobacter (0.15% vs 0.005%). Despite these subtle differences, the overall microbial compositions of CS and HS samples were not significantly different from each other (P < 0.05, Fig. 1), suggesting that home sterilization more effectively preserved the native HM microbiota than handwashing alone.

While overall community profiles did not differ significantly between HS and CS samples, HS samples exhibited a marked enrichment of skin-associated genera Bacillus and Propionibacterium (Table S4). Compared to CS samples, Bacillus was 350-fold more abundant in HS samples (mean difference: 6.97%; 95% CI 1.6%-12.3%), and Propionibacterium was 2.7-fold more abundant (mean difference: 5.70%; 95% CI − 2.4%-13.8%). Compared to WA samples, Bacillus was 233-fold more abundant (mean difference: 6.96%; 95% CI 1.5%-12.4%), and Propionibacterium was 8.1-fold more abundant (mean difference: 8.15%; 95% CI 1.1%-15.2%). These enrichments suggest that maternal skin bacteria may be introduced during manual reassembly of collection kits prior to pumping and may disproportionately shape relative abundance when total bacterial biomass is low.

Co-occurring pumping and hygiene practices and the microbiota of pumped milk

To capture real-world practices more accurately, we created composite variables based on the most commonly co-occurring hygiene and collection kit cleaning practices, which together accounted for 83.6% (n/N = 41/49) of observed behaviors. Specifically, 61% of participants expressed milk with unwashed hands using handwashed kits (UWA; n = 30); 22% used unwashed hands with home-sterilized kits (UHS; n = 11); Table S1). Each participant also expressed a second milk sample under study-controlled conditions using a hospital-grade pump and new, sterile collection supplies (STER supplies, denoted here as CS). These CS samples reflect “gold-standard” pumping practices and serve as a proxy for infant exposure during direct at-the-breast feeding. Each participants’ sterile sample serving as their own control, enabling direct assessment of real-world hygiene behaviors on the milk microbiota.

The phylogenetic diversity of 16S genes from UWA samples was significantly higher than that of CS samples (Faith’s PD, mean difference: − 8.3; 95% CI − 6.0 to − 11.2; P < 0.001), indicating that expressing milk with unwashed hands and handwashed kits was associated with a more diverse bacterial community than milk collected under sterile conditions that approximated direct at-the-breast feeding exposure. UHS samples exhibited intermediate diversity values that did not significantly differ from either UWA and CS samples (Fig. 3A), suggesting that home-sterilization mitigated the introduction of exogenous bacteria during pumping, even in the absence of pre-pumping handwashing.

Fig. 3.

Cleaning practices influence human milk microbiota structure and diversity among samples collected with unwashed hands (N = 93). A Boxplot of within-sample phylogenetic diversity (Faith’s PD), stratified by milk collection kit cleaning method: CS = commercially sterilized kits; UHS = unwashed hands and home-sterilized kits; UWA = unwashed hands and handwashed kits. Participants who washed their hands before pumping (n = 11) were excluded. Different letters indicate significant differences (P < 0.002). B Principal Components Analysis (PCA) of between-sample (beta) microbial diversity based on 16S ASVs, colored by cleaning method. Ellipses indicate expected location of new observations with 95% confidence (PERMANOVA). C Dendrogram and heatmap displaying microbial community structure at the genus level. Milk samples (columns) were grouped by complete linkage hierarchical clustering (CLC), which clusters samples based on a distance matrix derived from their genus-level relative abundance profiles. The resulting dendrogram reflects similarity in bacterial community structure, with clades labeled and visually separated by dashed lines in the heatmap. The two annotation rows beneath the dendrogram indicate each sample’s classification by (i) Dirichlet Multinomial Mixture (DMM) modeling, a probabilistic methods that assigns samples to compositional community types based on shared taxonomic patterns, independent of a distance matrix, and (ii) kit cleaning method. The heatmap (iii) shows the relative abundance of the 20 most abundant bacterial genera. Unid = Unidentified. Among milk samples collected without pre-pumping handwashing, differences in kit cleaning practices were associated with distinct microbial diversity patterns and taxonomic community profiles, as revealed by both beta-diversity and model-based classification

Taxonomic composition further supported these findings. UWA samples yielded significantly higher relative abundances of multiple genera of Proteobacteria genera compared to CS or UHS samples (all adjusted P < 0.05; Fig. 3C; Table S7; Fig. S1). Acinetobacter was 4.5-fold more abundant in UWA versus HS samples (mean difference: 13.07%; 95% CI 2.28%, 23.86%) and 105-fold higher than in CS samples (mean difference: 16.64%; 95% CI 7.28%, 26.0%). Stenotrophomonas was 78.9-fold higher than in HS samples (mean difference: 7.01%; 95% CI 0.46%, 13.56%) and 177.5-fold higher than in CS samples (mean difference: 7.06%; 95% CI 0.51%, 13.61%). Enterobacter was tenfold more abundant than in HS samples (mean difference: 1.35%; 95% CI − 0.59%, 3.29%) and 300-fold higher than in CS samples (mean difference: 1.5%; 95% CI − 0.42%, 3.41%).

Conversely, UHS samples yielded higher relative abundances of Bacillus and Propionibacterium compared to UWA or CS samples (all adjusted P < 0.05; Table S7; Fig. S1). Bacillus was 20.4-fold more abundant than in UHS versus UWA samples (mean difference: 7.00%; 95% CI − 3.90%, 17.90%) and 368-fold higher than in CS samples (mean difference: 7.34%; 95% CI − 3.88%, 18.56%). Propionibacterium was 18.0-fold higher in UHS versus UWA samples (mean difference: 10.01%; 95% CI − 4.73%, 24.75%) and 3.2-fold higher than in CS samples (mean difference: 7.29%; 95% CI − 5.87%, 20.45%).

Three independent multivariate analyses of β-diversity of corroborated these compositional differences (Fig. 3B: PCA, Fig. 3C: CLC, DMM, Tables S5 and S6). UWA samples clustered distinctly from CS and UHS samples (Fig. 3B, PERMANOVA R² = 0.36 overall P = 0.001; Fig. 3C), while microbial profiles of CS and UHS samples appeared more similar.

CLC and DMM models further substantiated the uniqueness of microbial profiles in UWA samples. Both algorithms identified three distinct milk community clusters (Fig. 3C). CLC Clade 1, comprising only of UWA samples, was characterized by dominance of Proteobacteria and corresponded to DMM group 3. In contrast, CLC Clades 2 and 3 (corresponding to DMM Groups 1 and 2) were, primarily composed of CS and UHS samples and were dominated by Streptococcus and Staphylococcus, respectively (Fig. 3; Table S7).

Culture-dependent analyses aligned with 16S sequencing-based observations of community structure. Samples in CLC Clade 1 yielded, on average, > 10⁴-fold higher gram-negative CFU/mL and elevated total aerobic bacterial counts compared to the other clades (adjusted P < 0.0001; Fig. 4). Gram-negative counts often exceeded 10⁵ CFU/mL in Clade 1, whereas median counts in Clades 2 and 3 fell below the limit of detection (~ 10 CFU/mL) (Fig. 4). Although clade assignment was based on hierarchical clustering of microbial composition and independent of reported hygiene practices, it closely aligned with them, particularly for Clade 1, which captured a distinct microbiological signature associated with the UWA combination. This group was significantly more likely to contain detectable and elevated levels of gram-negative bacteria than samples from other hygiene profiles (adjusted P < 0.0001; Fig. 4; Table S6), reinforcing the association between combined hygiene practices and microbial community structure.

Fig. 4.

Milk collected with unwashed hands into handwashed kits was more likely to contain culturable gram-negative bacteria (N = 93). Samples were grouped by complete linkage hierarchical clustering (CLC) of genus-level microbial composition. Clades reflect similarity in overall community structure. One clade (Clade 1) consisted exclusively of samples collected with handwashed kits and no pre-pumping handwashing (unwashed hands). This clade had significantly higher counts of gram-negative bacteria compared to Clades 2 and 3 (P < 0.0001), often exceeding 10⁵ CFU/mL. In contrast, the median counts for gram-negative bacteria in Clades 2 and 3 fell below the limit of detection (~ 10 CFU/mL). Not all UWA samples yielded gram-negative bacteria (Table S6). Different letters indicate significant adjusted differences P < 0.0001. Although clustering was performed independently of reported hygiene practices, the resulting clades aligned closely with them,particularly Clade 1, which captured a distinct microbial signature associated with the combination of handwashed kits and unwashed hands. This group was significantly more likely to contain detectable and elevated levels of gram-negative bacteria than samples from other hygiene profiles

In summary, our findings suggest that common real-world combinations of hygiene and cleaning practices may influence the microbial composition of expressed HM. Milk collected with unwashed hands and handwashed kits (UWA) was associated with increased bacterial diversity, a Proteobacteria-enriched profile, and elevated gram-negative bacterial loads. In contrast, milk expressed with unwashed hands and home-sterilized kits (UHS) more closely resembled CS samples in both community structure and bacterial load, with the exception of higher relative abundances of Bacillus and Propionibacterium. The robustness of results across culture-dependent and sequencing-based analyses strengthens confidence in these observations and highlights the potential impact of routine expression practices on the milk microbiota.

Discussion

This secondary analysis of the MiLC trial offers novel insights into how real-world, at-home pumping and hygiene practices shape the microbiota of expressed human milk. Among the factors examined, collection kit cleaning practices showed the strongest associations with both bacterial load and microbial composition, more so than personal breast pump type (open- vs. closed-system) or pre-pumping handwashing. Notably, the most common co-occurring combinations of hand hygiene and kit cleaning practices were also linked to distinct microbial profiles, underscoring the importance of investigating these behaviors in context rather than in isolation. While causal inferences are limited by the nature of this study design, our use of within-subject comparisons and multiple analytical methods strengthens the reliability of these findings and their relevance to milk microbial safety guidelines used by clinicians, milk banks, and millions of parents, [37].

Contrary to longstanding assumptions [38, 39], we observed no significant difference in bacterial counts between open- and closed-system personal breast pumps. This unexpected result may reflect improvements in breast pump designs or could have been confounded by other user behaviors, such as a lower frequency of handwashing observed among closed-system pump users. Other potential confounding factors include frequency of pump use and age of pump. The complexity of these interactions underscores the need for future studies to disentangle pump design effects from accompanying hygiene practices.

Surprisingly, only 1 in 5 participants washed their hands before pumping, despite CDC recommendations to do so [37]. While pre-pumping handwashing was not independently associated with statistically significant reductions in bacterial counts, the average total aerobic bacterial load in milk collected without pre-pumping handwashing exceeded 10⁴ CFU/mL, HMBANA’s threshold for “contamination” [36], whereas the average count in milk collected after handwashing did not. These findings, alongside prior evidence showing that hand hygiene instruction can reduce bacterial counts in milk [40], lend support to the current guidelines.

This study highlights the critical role of collection kit cleaning practices in shaping the microbiota of milk pumped at home. Although previous studies have reported associations between cleaning practices and levels of culturable bacteria on expressed milk in aggregate [41–45], our study extends these findings using both culture-based and 16S rRNA sequencing techniques. By comparing milk collected under real-world conditions (OWN) and milk collected using sterile techniques (STER), we were able to control for inter-individual variation and investigate the associations between individual and co-occurring hygiene and cleaning practices on the milk microbiota, offering a more comprehensive understanding than previous research.

Most MiLC participants reported washing their collection kits by hand with hot, soapy water in alignment widely promulgated guidelines, including those of the CDC [37, 43, 46]. However, this approach was associated with elevated counts of culturable gram-negative bacteria and a higher relative abundance of Proteobacteria, including genera such as Acinetobacter, Pseudomonas, and Stenotrophomonas, some of which include opportunistic pathogens [22, 47–51]. In contrast, milk collected using home-sterilized kits more closely resembled that of milk collected with hospital-grade pumps and sterile collection kits, representing the unaltered milk microbiota. These findings suggest that the microbiota of milk consumed from home-sterilized supplies is more similar to that of milk consumed by infants nursing directly at the breast.

Building on these findings, we also observed enrichment of skin-associated genera Bacillus and Propionibacterium in milk collected with home-sterilized kits (HS or UHS). These taxa were not similarly elevated in milk collected using handwashed kits (WA or UWA), suggesting that their overabundance is not a general marker of insufficient cleaning. Instead, it more likely reflects exogenous microbial input during the manual reassembly of pump parts. Both Bacillus and Propionibacterium are well-characterized members of the skin-associated microbiota and known contributors to HM via skin contact [9, 52, 53]. Their apparent overrepresentation in HS and UHS samples may be accentuated by the low total bacterial biomass of these samples, which increases their relative contribution even with modest exogenous inputs [54]. While these genera are not typically pathogenic in healthy individuals, certain species within the genera, particularly B. cereus, have been implicated in rare cases of infections in premature infants [55]. Due to the ability to survive pasteurization by both B. cereus and its heat-stable endotoxins, milk banks routinely screen for and discard donated milk containing B. cereus [44]. These findings emphasize that even when sterilization is employed, the subsequent handling and assembly of pump parts remains a critical control point for ensuring microbiological safety, particularly in high-risk feeding contexts.

At the same time, our findings highlight that manual washing of collection kits, when performed effectively, can also yield low bacterial counts in expressed milk. Several samples collected with handwashed collection kits yielded notably low concentrations of culturable total and gram-negative bacteria, suggesting that cleaning effectiveness appeared inconsistent across participants. One plausible explanation is variability in adherence to CDC guidelines. This hypothesis is supported by previous research by Carré et al. (2018), who reported that protocol deviations were prevalent despite standardized instruction, and such deviations were associated with elevated bacterial counts in expressed milk samples [56]. These findings raise the possibility of a disconnect between perceived and actual guideline adherence among pumping mothers. Additional research is needed into effective education strategies for hygienic pumping and cleaning practices. Such studies could provide valuable insights into optimizing at-home cleaning protocols.

A key differentiating factor in our study was the comparison of real-world pumping conditions against gold standard practices. Recognizing that multiple hygiene practices co-occur during milk expression, we developed composite variables to capture the most common practice combinations. Our findings, consistent across culture and 16S sequencing techniques, suggest that the combination of unwashed hands, paired with handwashed kits may introduce more exogenous bacteria into expressed milk than other practices observed, fundamentally altering its microbial composition. Importantly, our data suggest that adequate cleaning of collection supplies can significantly reduce the introduction of exogenous bacteria during pumping, even when hand hygiene is suboptimal. This corroborates previous evidence [41, 57, 58] and emphasizes the critical role of kit cleaning practices in shaping the expressed milk microbiota.

Our findings carry important implications for mothers pumping milk for hospitalized infants. Although our study focused on mothers of healthy infants, their pumping and hygiene practices, if used by mothers of hospitalized infants, would likely yield similar microbial profiles. Among milk collected using mothers’ own pumps, 50% contained culturable gram-negative bacteria, compared to only 5.8% of samples collected with sterile kits, a striking difference we previously reported [21]. These organisms produce lipopolysaccharide (LPS), a potent endotoxin that can trigger intestinal inflammatory and may increase the risk of NEC [59]. In addition to their inflammatory potential, gram-negative species include opportunistic pathogens that pose serious health risks to medically fragile infants [42, 49, 60, 61]. The gravity of this issue was highlighted in 2022, when bacterial contamination of milk pumped at home resulted in an infant death. That case prompted the CDC to revise its guidelines, now recommending sterilization of pump parts for hospitalized infants as well as healthy infants under two months of age [62, 63].

Although home sterilization is recommended to reduce bacterial load in expressed milk, several barriers may limit adherence among parents pumping milk for their hospitalized infants. Fatigue, stress, limited support, and logistical strain are common in the NICU context and can interfere with recommended hygiene practices [64, 65]. These barriers may lead to divergence from guidelines, such as improper cleaning (e.g., not disassembling kits), inadequate drying or storage conditions, and other deviations from protocol, even if standardized instructions are provided [56]. To address these challenges, interventions should pair clear institutional protocols with repeated, multimodal education, including verbal guidance, written instructions, illustrations, and hands-on demonstrations. Steam decontamination using microwave bags offers a simple and effective strategy, with evidence showing significant reductions in bacterial load [41]. In addition, multidisciplinary support from lactation consultants, nursing staff, and infection control teams, along with family-centered care models, can help mitigate both technical and psychosocial barriers [66–68]. Together, these strategies provide a framework for supporting parents in safely expressing milk for their vulnerable infants in the NICU.

The health implications of our findings for healthy infants, particularly those over 2–3 months old, remain uncertain. While we observed significant microbial changes in expressed milk, it is unclear whether these differences influence health outcomes in term infants. One possibility is that exposure to exogenous bacteria through pumped milk may have minimal impact in this population, given that older infants are already colonized and regularly exposed to environmental microbes. However, recent evidence suggests that the home environment itself shapes both the human milk and infant gastrointestinal microbiomes, with microbial sharing between mother, milk, and household surfaces contributing to colonization patterns [53, 67, 69].

Moreover, new studies suggest that milk microbes may play a more active role in gut development than previously understood. Shenhav et al. (2025) describe HM as a “microbial pacemaker” that regulates the timing of ecological transitions in the infant gut, influencing immune maturation and potentially shaping disease susceptibility [70]. Likewise, Noël-Romas et al. (2025) show that milk-origin microbes can persist in the gut and interact with other taxa well beyond the first month of life, particularly with sustained milk feeding [71]. These findings suggest that even in healthy, full-term infants, the composition of milk microbes, potentially influenced by hygiene practices and equipment use, may play a role in shaping the early-life gut ecosystem. While our study did not assess infant outcomes directly, these insights support the translational relevance of our findings and underscore the need for longitudinal studies.

A more pressing concern may be the potential effect of an elevated bacterial load on milk quality, thus effective storage duration. Human milk contains antimicrobial components such as lysozyme and lactoferrin that interact with bacteria [72], but their activity may be overwhelmed in samples with high initial bacterial counts. The Academy of Breastfeeding Medicine's milk storage guidelines suggest that expressed milk with higher bacterial loads may have a shorter shelf-life compared to milk with lower bacterial conts [43]. Our findings support this concern and highlight the need for further research on how bacterial load affects the milk storage duration. Such work could inform more tailored storage recommendations that preserve both the nutritional and immunological benefits of human milk.

This study had several notable strengths. A key strength of this study was the ability to compare milk expressed under real-world household conditions to milk collected using standardized, gold-standard procedures. By collecting milk using both OWN and STER collection supplies, participants served as their own controls. This allowed us to account for interindividual variability in ways that previous studies could not. Collecting milk in participants’ homes using their typical pumping practices, provided an accurate representation of infants’ microbial exposures through expressed milk. Finally, our use of both 16S rRNA sequencing and aerobic culturing techniques offered a more comprehensive assessment of the milk microbiota than traditional culturing alone [73, 74] and strengthened the relevance of previously reported associations between pumping practices and the microbial composition of pumped milk [19, 21, 41, 75].

Our study had some notable limitations, the most important of which was that participants were not randomized to pump cleaning methods or handwashing practices, thus, causal conclusions related to these factors cannot be made. RCTs with larger sample sizes are needed to confirm these findings. Additionally, we did not have strain-level data for our analyses, thus could not infer the presence of specific pathogens or the risk they might pose to infants. Longitudinal studies that assess the risk of illness as a function of milk microbiome dynamics with strain level resolution over time are clearly needed. Finally, a notable limitation of the present study was the absence of wearable breast pumps. These in-bra, hands-free breast pumps incorporate an integrated valve-membrane assembly and collection cup configuration that differs substantially from conventional breast pump collection kits [76]. Therefore, subsequent investigations should expand their scope to encompass these emerging technologies to provide a more comprehensive understanding of cleaning practices and milk microbial safety.

Conclusions

This study provides new insights into how common at-home milk expression practices shape the human milk microbiota. Across both culture-based and sequencing approaches, collection kit cleaning practices emerged as a more influential determinant of microbial composition of expressed milk than personal pump type or pre-pumping handwashing. Notably, most participants in our study did not wash their hands before pumping, and the most common hygiene combination, unwashed hands, paired with handwashed kits, was associated with the highest bacterial loads, enrichment of Proteobacteria, and distinct microbial community structures.

However, our findings also suggest that home-sterilization of milk collection kits may be one of the most effective ways to minimize the introduction of exogenous bacteria to milk during pumping. The resulting milk more closely resembled that from controlled sterile conditions, except for higher levels of skin-associated genera Bacillus and Propionibacterium highlighting the need for hygienic technique throughout the pumping process. Nevertheless, our findings suggest home-sterilized kits can approximate the microbial exposure of direct breastfeeding. These findings have important implications for milk microbial safety guidelines, particularly for milk intended for donation to milk banks or fed to critically ill infants, such as those born prematurely. While the health consequences of these microbial differences remain unclear for older, healthy infants, our results underscore the need for evidence-based pumping and hygiene guidelines that reflect real-world practices. Future research should prioritize longitudinal designs with strain-level resolution to evaluate the health effects of microbial exposures through pumped milk, and examine how bacterial load influences milk storage stability and safety.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

ASV

Amplicon sequence variant

CDC

Centers for disease control and prevention

CFU

Colony-forming units

CLC

Complete linkage clustering

CS

Commercially-sterilized

DMM

Dirichlet multinomial mixtures

FDR

False discovery rate

HM

Human milk

HMBANA

Human Milk Bank Association of North America

HS

Home-sterilized

LPS

Lipopolysaccharide

MiLC

Milk in life conditions

NEC

Necrotizing enterocolitis

OWN

Milk expressed with personal equipment

PCA

Principal components analysis

PERMANOVA

Permutational analysis of variance

SD

Standard deviation

STER

Milk expressed with hospital-grade, commercially sterilized equipment

UHS

Unwashed hands, paired with home-sterilized collection kits

UWA

Unwashed hands, paired with handwashed collection kits

WA

Washed

16S

16S ribosomal RNA

Author contributions

SMR involved in conceptualization, design, funding acquisition, conducting the study, having full access to data, taking responsibility for data integrity and accuracy, analyzing data, and drafting the initial manuscript. DLA involved in conducting culture experiments and providing administrative and technical support. JEW involved in supervising 16S rRNA analyses and providing administrative and technical support. MAM involved in supervising 16S rRNA analyses and interpreting data. MKM involved in interpreting data, providing feedback, and approving the final manuscript. KMR involved in conceptualization, design, funding acquisition, supervision, and interpreting data. AGH involved in conceptualization, design, funding acquisition, supervision, analyzing data, and drafting the initial manuscript. All authors participated in critical review and editing of the written drafts and approved the final manuscript.

Funding

This research was funded in part by NIH T32-DK007158 (SMR), USDA/Hatch NYC-399346 (KMR, AGH, SMR), a travel grant from International Society for Research in Human Milk and Lactation and Lactation and the Family Larsson-Rosenquist (SMR), McNair Scholars Program at Cornell University (DLA), NIGMS NIH P30 GM103324 (MAM), the Idaho Agricultural Experiment Station (MAM), and in-kind contributions from Medela, Inc. The funders had no role in the design and conduct of the study, decision to publish, or manuscript writing.

Data availability

All data underlying the findings described in this manuscript are freely available on NCBI Sequence Read Archive (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA1106482).

Declarations

Ethics approval and consent to participate

Written informed consent was obtained according to the study protocol approved by the Institutional Review Board at Cornell University (1608006566).

Consent for publication

Not applicable.

Competing interests

The authors have declared that no competing interests exist.