The Composition of Human Milk and Infant Faecal Microbiota Over the First Three Months of Life: A Pilot Study

Human milk contains a diverse array of bioactives and is also a source of bacteria for the developing infant gut. The aim of this study was to characterize the bacterial communities in human milk and infant faeces over the first 3 months of life, in 10 mother-infant pairs. The presence of viable Bifidobacterium and Lactobacillus in human milk was also evaluated. MiSeq sequencing revealed a large diversity of the human milk microbiota, identifying over 207 bacterial genera in milk samples. The phyla Proteobacteria and Firmicutes and the genera Pseudomonas, Staphylococcus and Streptococcus were the predominant bacterial groups. A core of 12 genera represented 81% of the microbiota relative abundance in milk samples at week 1, 3 and 6, decreasing to 73% at week 12. Genera shared between infant faeces and human milk samples accounted for 70–88% of the total relative abundance in infant faecal samples, supporting the hypothesis of vertical transfer of bacteria from milk to the infant gut. In addition, identical strains of Bifidobacterium breve and Lactobacillus plantarum were isolated from the milk and faeces of one mother-infant pair. Vertical transfer of bacteria via breastfeeding may contribute to the initial establishment of the microbiota in the developing infant intestine.

identified 14,15 . Interestingly, the study also reported that the human milk microbiota was compositionally distinct from other body niches and maternal weight and mode of delivery were reported as determinants of composition.
The origin of the human milk microbiome has been the subject of debate. Possible origins include bacteria from maternal skin and the oral cavity of the infant, where it has been demonstrated that during suckling, a high degree of retrograde flow back into the mammary ducts occurs 16 . More recently, the maternal gut has been suggested as a source, with bacteria entering the mammary glands via the entero-mammary pathway, a route that involves phagocytic dendritic cells penetrating the gut epithelium and trafficking bacteria through the circulatory system [17][18][19] . This is important since manipulation of the maternal microbiota could be used to promote an optimal human milk microbiome.
Given the significant health effects that breast-feeding exerts on infant health and development, it is important to define the composition of the human milk microbiota. Accordingly, the aim of this study was to characterise the microbiota composition of human milk in ten lactating women, and correlate the findings with gut microbiota composition of their infants, over the first three months of life. Additionally, we assessed the presence of culturable Bifidobacterium and Lactobacillus strains and their recovery from the infant gut.

MiSeq Sequencing of Human Milk and Infant Faeces. MiSeq sequencing of human milk and infant
faecal samples yielded a total of 20,206,055 reads, with mean read lengths of 460 bp (330-591 bp). Following quality control, an average of 61,786 (range 34836-109067) and 189,861 (range 98772-327481) reads per sample were obtained for human milk and infant faeces, respectively. Reads were classified into 1313 observed operational taxonomic units (OTUs) for human milk and 264 OTUs for infant faeces at a 3% similarity cut-off. All samples were rarefied to 33,000 sequences to prevent bias due to sampling depth.
To estimate microbial richness we used the Chao1 richness estimator which revealed significantly higher bacterial richness in milk samples than faecal samples at week 1, 3, 6 and 12 (p < 0.0001, p < 0.0001, p = 0.0002 and p = 0.0003 respectively) ( Fig. 1). We applied the Simpson's index to estimate microbial evenness and this was significantly higher in milk samples at week 1 and 6 (p = 0.002 and p = 0.004). To predict microbial diversity we used the Shannon index, which combines both bacterial richness and evenness, and is more responsive to rare species in terms of species richness than the Simpson's index which is more sensitive to dominant species. The Shannon index was significantly higher in milk samples at week 1, 3, 6 and 12 (p = 0.0012, p = 0.005, p < 0.0001 and p = 0.004). Over time, there was a significant increase in bacterial richness, Chao1, from week 3 to 6 (p = 0.0462) and a significant decrease in richness from week 6 to 12 (p = 0.0462) in milk samples. The Shannon index demonstrated a significant decrease in diversity in milk samples from week 6 to 12 (p = 0.0482). There were no statistically significant differences in alpha diversity metrics in fecal samples over time.
At genus level, 12 genera appeared to predominate in human milk, as they were detected at a mean relative abundance of ≥ 1% in at least 90% of samples collected over time (Fig. 3). These genera dominated the community, representing 81% of the relative abundance in human milk samples at week 1 (range 64-91%), 3 (range 64-90%) and 6 (range 61-90%) and 73% of the relative abundance at week 12 (range 50-82%). This core comprised Pseudomonas, Staphylococcus, Streptococcus, Elizabethkingia, Variovorax, Bifidobacterium, Flavobacterium, Lactobacillus, Stenotrophomonas, Brevundimonas, Chryseobacterium and Enterobacter. The remaining sequences mapped to 195 genera, further demonstrating the diversity of the human milk microbiota (Supplementary Table S1). The relative abundances of these genera were individual specific and subject to intra-individual variations over time.
There was little temporal stability in milk samples from most lactating women, as the relative abundance of the bacterial genera present shifted over time (Supplementary Table S1). With the exception of week 3 where Streptococcus had the highest mean relative abundance (32%), Pseudomonas was predominant at all other sampling points detected at 21%, 27% and 19% at week 1, 6 and 12 respectively. By contrast, only 50 genera were detected in infant faecal samples. Pseudomonas was not detected by sequencing and the samples were dominated by Staphylococcus at week 1 (19%), Escherichia-Shigella at week 3 and 12 (17% and 21% respectively) and Veillonella at week 6 (23%) (Fig. 4). When considering genera detected in both human milk and faeces (Table 1), these accounted for 88%, 85%, 88% and 70% of the total reads in infant faecal samples at week 1, 3, 6 and 12 respectively. For human milk, these genera accounted for 37%, 51%, 19% and 27% of the total reads. These genera occurred with varying relative abundances and frequencies. Veillonella and Escherichia/Shigella for example, occurred at a higher mean relative abundance in infant faeces, 14% and 12% respectively, compared to 1% and 0.04% respectively in human milk at week 1. In contrast, Staphylococcus was consistently detected at a high frequency in all samples with a mean relative abundance of 13% and 19% at week 1 in human milk and infant faeces, respectively.
A number of niche-specific genera were exclusive to either human milk or infant faeces. For example, the high relative abundance of Proteobacteria and Bacteroidetes in human milk was largely attributable to the genera Pseudomonas and Variovorax, and Elizabethkingia and Flavobacterium respectively, which were exclusive to human milk. In infant faeces, Eggerthella, which was the only genus not detected in human milk samples, contributed to the higher relative abundance of Actinobacteria.
One subject, M10, reported symptoms of subacute mastitis and withdrew from the study at week 6 following antibiotic administration. Among the milk samples obtained from this subject, the relative abundance of Staphylococcus was higher than the mean relative abundance for healthy subjects, accounting for 73% and 24% Scientific RepoRts | 7:40597 | DOI: 10.1038/srep40597 of the reads obtained at week 1 and week 3, respectively, compared to 12% and 6% in healthy women. Alpha diversity metrics were also lowest in the mastitic milk, this likely reflects the low abundance of other species in this sample. Also of note was mother-infant pair 9, where Haemophilus was detected at a relative abundance of 24% in the human milk sample and 33% in the infant fecal microbiota at week 1. The mean relative abundance for Haemophilus in other human milk samples at week 1 was 0.1% and 0.5% in infant faeces.
Principal coordinate analysis (PCoA) plots using weighted UniFrac distances demonstrated a clear separation of milk and faecal samples at each week (Supplementary Figure 1). There were no distinct clusters of milk samples over time, although bacterial communities clustered more closely to one another at week 12 than at week 1 (Supplementary Figure 2a). Similarly, in faecal samples no obvious separation was observed between samples at different weeks (Supplementary Figure 2b).

Culture-Dependent Analysis of Human Milk and Infant Faeces.
Culture on selective media revealed the presence of presumptive Bifidobacterium and Lactobacillus in the human milk of one mother at 1 × 10 2 CFU/ml and 3 × 10 3 CFU/ml respectively at week 3. Bifidobacterium and Lactobacillus were isolated from the corresponding infant fecal sample at 4 × 10 6 CFU/ml and 7 × 10 7 CFU/ml, respectively. No culturable Bifidobacterium or Lactobacillus were detected in human milk samples from other mothers in the study. Following enumeration, bacterial isolates from the human milk and the corresponding infant stool sample were subjected to 16S rRNA gene sequencing using species-specific primers. This identified the Bifidobacterium isolates as B. breve and Lactobacillus isolates as L. plantarum in both the human milk and the infants fecal samples. In addition, PFGE analysis of these isolates revealed identical profiles in both the human milk and the infant fecal samples for both B. breve and L. plantarum, which indicated that these isolates belonged to the same strain (Fig. 5).

Discussion
The benefits of human milk in terms of infant health and development have been well documented 1,8,10,12,17,20 . Human milk has been recognized as a fundamental source of bioactive components including bacteria that may contribute to neonatal gastrointestinal colonisation and immune development and maturation during the crucial early stages of development [8][9][10] . Differences have been reported in the microbiota composition of breast-fed infants versus formula-fed with the former suffering from less allergies and gastrointestinal infections 3,4 . Therefore, the microbiota of breast-fed infants is considered the gold standard in terms of a healthy infant gastrointestinal microbiota. Comprehensively characterising the human milk microbiota is vital for enabling better insight of its significance and activity in relation to the developing infant gut microbiota and health.
In this study, Illumina MiSeq sequencing revealed that 12 genera; Pseudomonas, Staphylococcus, Streptococcus, Elizabethkingia, Variovorax, Bifidobacterium, Flavobacterium, Lactobacillus, Stenotrophomonas, Brevundimonas, Chryseobacterium and Enterobacter dominated the milk of most lactating women, constituting a core milk microbiota. The presence and relative abundances of the remaining 195 genera were unique to individual mothers and subject to variation over time. The core microbiota constituted 81% of the taxa identified at week 1, 3 and 6, however by week 12, this decreased to 73%, suggesting a selected core microbiota drives the early stages but individual-specific taxa become more important in later stages of lactation.
Pseudomonas has been reported as a dominant member of the human milk microbiota in several studies including Ward et al., where it accounted for 61% of the relative abundance from milk samples taken between 9 and 30 days postpartum 14,21,22 . Similarly Staphylococcus has been found to be a common constituent of the milk microbiota by both culture independent and dependent investigations 11,14,15,[21][22][23] . In a study of breast-fed Swedish infants, both vaginally and caesarean section delivered, 100% of infant feces were colonised by Staphylococcus from day 3 of life. Staphylococcus epidermis in particular appears to have a biological relevance as it has been shown to be the predominant species in human milk and in the faeces of breast-fed infants and is less common in stool of formula-fed infants 11,24,25 . Genera associated with the oral cavity such as Streptococcus and Veillonella' reported by Hunt et al. and Cabrera-Rubio et al. were also prevalent in this study 14,15 . Although not previously reported in the human milk microbiome, Variovorax, strains of which have been isolated from the human oral cavity, were consistently detected in all samples 26 . The detection of large proportions of typical inhabitants of the skin and oral microbiota may imply that the origin in this case is secondary contamination. However, anerobic gut-associated populations such as Bacteroides, Blautia, Faecalibacterium, Ruminococcus, Roseburia, Subdoligranulum, Enterococcus and Escherichia-Shigella were also detected here and in other studies 15,21 . Bifidobacterium was also consistently detected, supporting the findings of Jost et al., and Hunt et al., however Cabrera-Rubio et al. and Ward et al. did not report the presence of Bifidobacterium. These differences are likely due to the method of DNA extraction used, as a bead-beating step was not incorporated in the latter studies. When making comparisons across studies, it is important to note that differences in microbial community composition may also have been affected by factors such as the hypervariable region of the 16S gene examined (we targeted the V3-V4 hypervariable regions of 16S rDNA), geographical differences and the greater sequencing depth achieved using Illumina MiSeq sequencing here compared with Roche 454 pyrosequencing used in other studies. These factors are known to influence diversity and richness estimates and can greatly impact the microbiome of individuals.
When considering the infant fecal microbiota, it was found to be less diverse than that of human milk. The communities were most similar with respect to Staphylococcus which accounted for a mean relative abundance of 15% and 19% at week 1 in human milk and infant faeces, respectively. A number of typically gut-associated genera were common to both human milk and infant faeces including Bifidobacterium, Bacteroides, Enterococcus, Lactobacillus, Clostridium, Coprococcus, Escherichia-Shigella and members of the Lachnospiraceae family. Interestingly, these shared genera accounted for 70-88% of the total reads in infant faecal samples throughout the sampling period. This is in agreement with other studies in mother-infant pairs which have shown that the bacterial composition of the faecal microbiota of the breast-fed infants reflects that found in the breast milk 12,24,[27][28][29][30] . In these studies, the genera Lactobacillus, Staphylococcus, Enterococcus and Bifidobacterium were frequently shared between breast milk and infant faeces. Jimenez et al., in a study with 23 mother-infant pairs identified Staphylococcus as the predominant species in milk and breast-fed infants faeces 25 . Gronlund et al., reported that maternal breast-milk bifidobacterial counts impacted on the infants' faecal Bifidobacterium levels 27 . While Jost et al., identified a number of gut-associated anaerobic genera like Bifidobacterium, Bacteroides and members of the class Clostridia shared between milk and infant faeces 30 .
Stronger evidence of vertical transfer from mother to infant involves identification of identical strains. Lactobacillus and Bifidobacterium are often considered as members of a healthy microbiota with the predominance of Bifidobacterium in particular appearing to be characteristic of the healthy breast-fed infant and were therefore targeted for strain identification. Despite being detected by MiSeq sequencing, culturable Lactobacillus and Bifidobacterium were found in only one milk sample with a low bacterial count of 3 × 10 3 CFU/ml and 1 × 10 2 CFU/ml, respectively. Similarly, Albesharat et al., were unable to isolate any bifidobacteria from human milk samples 29 . It is unclear whether this was due to their low abundance in human milk, the presence of antimicrobial compounds, the length of storage time of samples prior to culturing and in the case of bifidobacteria, their anaerobicity i.e. viable but not culturable. As MiSeq sequencing does not distinguish between live and dead cells, it is also possible that dead cells are being transferred during feeding which would nonetheless elicit an immune response. Of interest in this case was that these culturable isolates belonged to the same species as those isolated from the corresponding infant fecal samples, namely L. plantarum and B. breve. PFGE analysis confirmed that the same bacterial strain was shared in this mother-infant pair supporting the notion of vertical transfer via human milk. Other studies have also isolated the same Bifidobacterium and Lactobacillus strains from human milk and infant faeces including strains of B. breve and L. plantarum 13,28 .
A limitation of this study is the relatively small sample size and further investigations in larger populations, including maternal faecal samples are planned to confirm these results and extend the knowledge about the milk microbiome. In summary, our data demonstrate the large diversity of the human milk microbiota with over 207 bacterial genera identified. The relative abundances of these were unique to each mother and subject to variation over time. Coupled with our finding that the same strains of Bifidobacterium and Lactobacillus were found in maternal human milk and corresponding infant fecal samples, these results suggest that there is a microbiota specific for each mother-infant pair that could confer benefits specific to that infant. This is particularly pertinent as commercially available infant formulas and donor milk are sterilized/pasteurised and as such contain little to no microbes. The results of this work also confirmed the presence of microbes typically associated with the gut microbiota in milk samples suggesting strategies to manipulate the maternal gut with bacteria which could confer benefits to the infant may be beneficial. These data emphasise that human milk constitutes a relevant source of a wide range of bacteria for the infant gut and can contribute to infant gut colonisation and therefore to infant health.

Subjects, Study Design and Sample Collection. This study was approved by the Clinical Research
Ethics Committee of the Cork Teaching Hospitals. Parents of infant participants provided written informed consent and all relevant guidelines and regulations were followed. Ten mother-infant pairs were recruited at Cork University Maternity Hospital. Recruits were healthy lactating women and their full-term, healthy breast-fed infants (breast-fed for a minimum of 4 weeks after birth). The clinical characteristics of mother-infant pairs are shown in Table 2. Milk and faecal samples were collected from each mother-infant pair at 1, 3, 6 and 12 weeks. For milk sampling, sterile gloves were worn and the nipple and areola of the human were cleaned with chlorhexidane wipes (Clinell, United Kingdom) prior to manual expression of human milk into a sterile tube. The first few drops (approximately 1 ml) were discarded to prevent chlorhexidine contamination. Faecal samples were also collected into sterile containers and stored at 4 °C until delivery to the laboratory. Aliquots of 1 ml of fresh milk and 1 g Sequence and Statistical Analysis. 300 bp paired-end reads were assembled using FLASH with parameters of a minimum overlap of 20 bp and a maximum overlap of 120 bp 31 . The QIIME suite of tools, v1.8.0, was used for further processing of paired-end reads, including quality filtering based on a quality score of > 25 and removal of mismatched barcodes and sequences below length thresholds 32 . Denoising, chimera detection and operational taxonomic unit (OTU) grouping were performed in QIIME using USEARCH v7 33 . Taxonomic ranks were assigned by alignment of OTUs using PyNAST to the SILVA SSURef database release 111 34,35 . Alpha and beta diversities were generated in QIIME and calculated based on weighted and unweighted Unifrac distance matrices 36 . Principal coordinate analysis (PCoA) plots were visualised using EMPeror v0.9.3-dev 37 . To determine any statistically significant differences in microbial diversity between milk and faecal samples, non-parametric Mann-Whitney analysis was completed using Minitab 15 statistical software package. Statistical significance was accepted as p < 0.05, adjusted for ties.
Isolation and Enumeration of Bifidobacterium and Lactobacillus spp. Aliquots of faecal sample,