Human milk microbiota profiles in relation to birthing method, gestation and infant gender

1With the incidence of various non-infectious diseases on the rise, there is much interest in the developmental origins of health and disease and the potential role of early life feeding practices in modulating these outcomes. Breast-fed infants have been shown to be better protected than formula-fed infants against necrotizing enterocolitis and diarrhoea, allergy and asthma, inflammatory bowel disease, type 1 and type II diabetes, obesity and cardiovascular disease. In addition to immune protection and bioactive compounds being conveyed through maternal milk, a possible protective role of bacteria has been suggested. Lower than average levels of Bifidobacterium in human milk correlate with low levels of Bifidobacterium in the neonatal gut, allowing for higher than normal levels of Bacteroides to be established. These high levels of Bacteroides early in life have been associated with an increased risk of asthma and obesity later in life. Indeed, efforts to manipulate the microbiota of formula-fed infants through probiotic supplementation have resulted in protection against some of the above diseases, comparable to that observed for breast-fed infants.

2Differences exist in bioactive components, macronutrients, cytokines, enzymes, proteins and immunological factors between preterm and term milk and milk from mothers giving birth by vaginal and caesarean deliveries. As well, the energy content differs in milk depending on gender of the newborn, with breast milk from mothers who give birth to sons having more fat content than that of daughters.

3We hypothesized that physiological or hormonal triggers that influence milk composition might also support different bacterial genera. Thus, we studied human milk samples from women giving birth at different stages of gestation, by vaginal or caesarean delivery, and examined whether gender of the newborn also affected the microbiota profiles.

4Characterization of the milk microbiota in 39 Canadian women showed that despite diverse clinical parameters, almost all had high abundances of Staphylococcus, Enterobacteriaceae and Pseudomonas in their milk (Fig. 1) (Additional file 1: Table S1). Overall, the top 6 most abundant taxa represented in milk were Staphylococcus (31 %), Enterobacteriaceae (10 %), Pseudomonas (17 %), Streptococcus (5 %) and Lactobacillus (3 %) (Additional file 2: Figure S1). Out of the 47 genera detected, half (51 %) were present in all samples (Additional file 3: Table S2). At the phylum level, Proteobacteria and Firmicutes made up the highest proportion of phyla in all samples with Actinobacteria and Bacteroidetes also present but at lower levels (Fig. 2).

5No differences were detected in microbial profiles based on gestation, mode of delivery or gender of the child, using both weighted (Fig. 3a) and un-weighted UniFrac distances (data not shown). The above two matrices, however, assign a large emphasis on either rare or highly abundant taxa, so compositional changes that occur in moderately abundant lineages may be overlooked. For this reason, we also analyzed the data using generalized UniFrac at an alpha of 0.5, which overcomes this problem. As with UniFrac, no differences were observed (Fig. 3b), but there was separation of 13 samples, forming 2 distinct clusters, which could not be explained by any of the other metadata that was collected (Additional file 4: Figure S2) (Additional file 1: Table S1). To further confirm the above results, we also utilized the Bray-Curtis dissimilarity metric, which does not take into account phylogenetic relatedness of the biological community, as does UniFrac. As expected, no differences were observed (Additional file 5: Figure S3).

6The dataset was also analyzed using ALDEx2 to examine whether specific taxa were differentially expressed based on gestation, mode of delivery or gender. Again, no differences were detected at the genus (Additional file 6: Table S3), family, class or phyla levels (data not shown).

7Of interest, the abundance profiles clearly showed that mothers are transmitting very different bacterial profiles to newborns. For example, the milk of subject 40 contained over 80 % abundance of staphylococci, whereas that of subject 25 comprised mostly Gram-negative organisms, especially Pseudomonas.

8This study revealed a range of bacterial genera in human milk, consistent with previous studies. Interestingly, even when a baby was born extremely prematurely (subject S24), the mother’s milk was similar in composition to that of a woman giving birth at full term (subject S30). It can be speculated that this might be a fail-safe mechanism whereby the mother is “ready” to pass along her bacterial imprint irrespective of when the baby is born. If so, the microbiota would appear to be recalcitrant to late pregnancy hormonal and inflammatory changes, which could indicate an evolutionary pressure protecting this niche for the baby’s benefit. Further studies on this concept are warranted, and cases where the milk profiles are very different or the outcome for the baby negative could be particularly insightful.

9It was not surprising that milk from emergency caesarean (C) section deliveries (i.e. non-elective) did not differ from women who gave birth vaginally, as this decision is made at the time of labour, when the hormones and timing for birth have been initiated. We also did not see differences between non-elective C section deliveries and vaginal births which is in contrast to a study published by Cabrera-Rubio et al. in which they concluded that the human milk microbiome is shaped by mode of delivery. However, we do not believe that the authors adequately proved this claim. In their analysis, the milk samples collected were from both obese and normal weight women with no indication of the proportion in each group. Since it was shown in that same paper that body mass index influences the milk microbiota, the subject’s weight could have been a confounding factor responsible for the observed differences. In addition, the author’s claim that the milk microbiota is influenced by mode of delivery was based solely on visual observations of bar plots (which were not very distinct between the two groups), with no principal coordinate analysis (PCoA) plots or statistical analyses to support this claim. These conflicting results between our group and that of Cabrera-Rubio highlight the need for future studies with larger sample sizes and inclusion of women from various demographics.

10Because of the multivariate nature of the data, differences, if they are present, may have been masked by the different variables confounding each other. Thus, a larger sample size allowing for linear regression analysis would strengthen our claims. However, there is the possibility that no matter the sample size, differences will only be apparent when examined at the level of the individual. Schwarzberg et al. showed that treatment for periodontitis had no effects on microbial profiles when compared to controls. However, when pre- and post-treatment effects for each individual were compared, significant changes in bacterial composition were observed. The authors thus emphasized the importance of comparing shifts from a personalized healthy state to a personalized disease state in order to truly understand biological changes.

11Proteobacteria and Firmicutes were the dominant phyla, consistent with a previous high-throughput study performed in Switzerland and analyzed with a different sequencing platform (454 sequencing).

12The detection of Staphylococcus, as the most abundant organism, is likewise consistent with other studies. Staphylococcus is more abundant in the gut of breast-fed newborns compared to formula-fed ones, but their numbers start to decrease after the first week of life when oxygen has been consumed and an environment favourable for anaerobes is created. Unlike Staphylococcus, which is present in high abundance in at least the first week of life, Proteobacteria are never found in high numbers in the faeces of newborns, infants or Western adults. Thus, an important question arises, as to what the evolutionary significance is of having such a diverse population of bacteria in milk, if only a select few seem to colonize during the development of the neonate. There are a few possibilities; firstly, persistent colonization is not always needed for beneficial effects, as transient exposure can be just as effective. Secondly, bacteria in milk may not have to be passed on from the mother to the child to exert their beneficial effects. Many of the protective factors in milk such as antibodies, immune cells, lactoferrin and beta defensins originate from the mammary gland and not from the blood. With the ability of bacteria to regulate host gene expression, such as anti-microbials, and their ability to stimulate the immune system, the plethora of bacteria in breast milk could be inducing up-regulation of these protective factors in the breast that then get passed on in high concentrations to the neonate via milk.

13From another perspective, there is the possibility that some of these milk microbes have either limited or no effects on the offspring but are present for the benefit of the mother, such as in the protection against mastitis. Mastitis is a painful inflammatory condition of the breast with the main causative agent being Staphylococcus aureus, and it has been shown that some milk commensals have the ability to prevent S.aureus growth and infection. The same is true for a skin derived strain of Propionibacterium acnes, which prevents growth of S. aureus via its by-products of glycerol fermentation. With the high abundance of glycerol present in milk, milk-derived strains of Propionibacterium may have similar growth-inhibiting properties, which would account for its presence in every milk sample collected.

14Lactobacillus was present in high abundance in milk, so for those women who deliver by C section and thus do not transfer lactobacilli from the vagina to the infant, the milk could provide a means for these beneficial organisms to reach the infant gut.

15A diverse population of bacteria is present in breast milk dominated by the phyla Proteobacteria and Firmicutes and the taxa Staphylococcus, Pseudomonas, Enterobacteriaceae, Streptococcus and Lactobacillus. While no differences in microbial profiles were apparent based on gestation, mode of delivery or gender, more studies are still needed on what factors do influence bacterial communities in milk and how these changes impact neonatal and maternal health.

16A single milk sample (day 6 and onwards post-partum) was collected from 39 Caucasian Canadian women recruited from London, Ontario, and the surrounding area, representing a homogenous community. Even though the samples collected were from different days after birth, with some considered “transitional” and others “mature,” a Kendall’s tau correlation test showed no statistically significant correlation between operational taxonomic unit (OTU) relative abundances over time after a Benjamini-Hochberg correction for multiple hypothesis tests. This shows that it is acceptable to use all milk samples, regardless of when it was collected, in our analyses. Ethical approval was obtained from Western Research Ethics Board and Lawson Health Research Institute, London, Ontario, Canada. Subjects provided written consent for sample collection and subsequent analyses. Participants were excluded if they were suffering from mastitis and were/had been on antibiotics during lactation. Caesarean deliveries were classified as either (i) “non-elective”, if there were complications during labour, or (ii) “elective”, if they were planned in advance or if the health of the foetus and/or mother was at risk prior to labour.

17Wearing sterile gloves, the women cleaned their nipple and surrounding area with sterile saline swabs to reduce the presence of skin bacteria. Milk was collected using a sterile HygieniKit Milk Collection System (Ameda, Buffalo Grove, IL, USA) attached to an electric breast pump. Between 5 and 15 ml of milk was pumped into a sterile tube and kept on ice until transfer to the laboratory, which occurred within 1 h of collection. Samples were aliquoted and stored at −20 °C until DNA extraction.

18After thawing on ice, 2 ml of milk were spun down at 20,000×g for 10 min and the supernatant was discarded. The pellet was then homogenized in 1.4 ml of ASL buffer (QIAamp® DNA Stool Kit, QIAGEN: Valencia, CA, USA) and 400 mg of 0.1-mm diameter zirconium-glass beads (BioSpec Products, Bartlesville, OK, USA). Mechanical and chemical lyses were performed by bead beading at 4800 rpm for 60 s, then 60 s on ice (repeated twice) using a mini-beadbeater-1 (BioSpec Products) and then incubated at 95 °C for 5 min. Subsequent procedures were performed using the QIAGEN QIAamp® DNA Stool Kit according to the manufacturer’s protocol, with the exception of the last step in which the column was eluted with 120 μl of elution buffer. DNA was stored at −20 °C until further use.

19A no template PCR control and a DNA extraction kit reagent control were sequenced alongside the samples. We observed that the taxon abundances in the controls were uncorrelated with the abundances in the experimental samples, and the distance between the controls and samples was large. Thus, we conclude that the controls had different profiles than that of the milk samples (Additional file 7: Figure S4).

20The genomic DNA isolated from the clinical samples was amplified using barcoded primers that amplify the V6 hypervariable region of the 16S ribosomal RNA (rRNA) gene which is 70 base pairs long:V6-Forward: 5′ACACTCTTTCCCTACACGACGCTCTTCCGATCTnnnn(8)CWACGCGARGAACCTTACC 3′V6-Reverse: 5′CGGTCTCGGCATTCCTGCTGAAC CGCTCTTCCGATCTnnnn(8)ACRACACGAGCTGACGAC 3′

21V6-Forward: 5′ACACTCTTTCCCTACACGACGCTCTTCCGATCTnnnn(8)CWACGCGARGAACCTTACC 3′

22V6-Reverse: 5′CGGTCTCGGCATTCCTGCTGAAC CGCTCTTCCGATCTnnnn(8)ACRACACGAGCTGACGAC 3′

23nnnn indicates four randomly incorporated nucleotides and “8” represents a specific sample barcode sequence. The PCR was carried out in a 42 μl reaction containing 2 μl of DNA template (or nuclease-free water as a negative control), 0.15 μg/μl bovine serum albumin, 20 μl of 2X GoTaq hot start colourless master mix (Promega) and 10 μl of each primer (initial concentration 3.2 pmol/μl). Thermal cycling was carried out in an Eppendorf Mastercycler under the following conditions: initial denaturation at 95 °C for 2 min followed by 25 cycles of 95 °C for 1 min, 55 °C for 1 min, and 72 °C for 1 min. After amplification, the DNA concentration was measured with the Qubit® 2.0 Fluorometer (Invitrogen) using the broad range assay. Equimolar amounts of each PCR product were then pooled together and purified using the QIAquick PCR purification kit (QIAGEN). The pooled PCR purified sample was then paired end sequenced on the Illumina MiSeq platform using a 150-cycle kit with a 2 × 80 run at the London Regional Genomics Centre, London, ON, Canada, following standard operating procedures.

24Custom Perl and Bash scripts were used to de-multiplex the reads and assign barcoded reads to individual samples. Multiple layers of filtering were employed: (i) Paired end sequences were overlapped with Pandaseq, allowing zero mismatch in the overlapped reads; (ii) Reads were kept if the sequence included a perfect match to the V6 16S rRNA gene primers; (iii) Barcodes were 8mers with an edit distance of >4, and reads were kept if the sequence were a perfect match to the barcode; (iv) Reads were clustered by 97 % identity into OTUs using the Uclust algorithm of USEARCH v7 which has a de novo chimera filter built into it; and (v) All singleton OTUs were discarded, and those that represented ≥1 % of the reads in at least one sample were kept.

25Taxonomic assignments for each OTU were made by extracting the best hits from the SILVA database and then manually verified using the Ribosomal Database Project (RDP) SeqMatch tool (rdp.cme.msu.edu) and by BLAST against the Green genes database (greengenes.lbl.gov). Taxonomy was assigned based on hits with the highest percentage identities and coverage. If multiple hits fulfilled this criterion, classification was re-assigned to a higher common taxonomy. A summary of each OTU classification and its sequence is shown in Additional file 8: Table S4. The raw sequencing reads generated in this study have been deposited to the NCBI Short Read Archive (SRA) database accession # SRP 064311.

26Weighted and un-weighted UniFrac distances were calculated in QIIME by using a tree of OTU sequences built with FASTTREE based on an OTU sequence alignment made with MUSCLE. The QIIME pipeline was also used to generate PCoA plots to visualize the Bray-Curtis dissimilarity. Changes in the microbial community composition were also analyzed by calculating the generalized UniFrac distance (alpha = 0.5, rooted phylogenetic tree) using the GUniFrac package in R, version 3.1.2. PERMANOVA was used to test for statistical significance between the groups using 10,000 permutations (QIIME package). Bar plots, box plots, stripcharts, PCoA plots and k means clustering analysis were all generated in R (http://www.R-project.org/).

27The ALDEx R package version 2 was used to compare genera, class, family and phyla between preterm and term milk; caesarean and vaginal deliveries, and male and female children. Values reported in this manuscript represent the expected values of 128 Dirichlet Monte Carlo instances. A value of zero indicates that organism abundance is equal to the geometric mean abundance. Thus, organisms more abundant than the mean will have positive values, and those less abundant than the mean will have negative values. Base 2 was used for the logarithm, so differences between values represent fold changes. Significance (p < 0.05) was based on the Benjamini-Hochberg corrected p value of the Welch’s t test and the Wilcoxon test.

28The data set supporting the results of this article are included within Additional file 1: Table S1; Additional file 2: Figure S1; Additional file 3: Table S2; Additional file 4: Figure S2; Additional file 5: Figure S3; Additional file 6: Table S3; Additional file 7: Figure S4; Additional file 8: Table S4.