Mother-to-Infant Microbial Transmission from Different Body Sites Shapes the Developing Infant Gut Microbiome

1The complex microbial community that inhabits humans, the microbiome, is an integral aspect of human health. In what is undoubtedly a complex interplay between host genetics and environmental conditions, these resident microbes support many functions in the human body, including the facilitation of nutrient absorption that would otherwise be inaccessible to the host, the training and modulation of the immune system, and protection against pathogenic organisms. Dysbiosis of this harmonious relationship has reported to be linked to many diseases in adults, including inflammatory bowel diseases (IBDs), type 2 diabetes, and colorectal cancer. Similarly, in infants it is associated with IBD, Crohn's disease, type 1 diabetes, necrotizing enterocolitis, and asthma. While the importance of the host-microbiome interplay is not in question, the mechanisms by which an infant acquires these microbes, and from what source, remain largely unexplored.

2The long-held belief that an infant is sterile at birth has been challenged by an increasing number of reports offering evidence of the occurrence of intra-uterine seeding, but the role and importance of prenatal microbial colonization are still open to debate. What is clearer is that extensive microbial colonization begins postpartum. Several crucial factors have been linked to the early development of the infant microbiome, including the mode of delivery and gestational age at birth, as well as other influencing factors including maternal and infant antibiotic usage and feeding method (formula or breastfeeding). Wider environmental exposure and early intimate relations, particularly with the mother, also play a pivotal role in the early microbial acquisition and community succession in the infant.

3Much of what is known about the acquisition of the early infant microbiome has been obtained using cultivation-based or taxonomic profiling limited to a species-level resolution. As individuals frequently share common species, inferring transmission at lower taxonomic resolutions, even at the species level, is not sufficient as a species can comprise multiple subspecies strain variants, which can be specific to different individuals. It is essential, therefore, to use strain-level profiling to identify and quantify the instances of transmission from external sources to the infant. This has been shown for a limited number of cultivable species, and we previously demonstrated that the maternal microbial reservoir is an important source in the early acquisition of microbial species and strains in the infant gut. Yet there has been no comprehensive assessment of the multiple potential maternal sources of microbial transmission, and how ultimately they contribute to the acquisition of the infant microbiome within hours of birth and over the first few months of life.

4To this end, we investigated what is arguably the most important intimate relationship in the development of the early infant microbiome, the mother. Focusing on five potential maternal sources of microbial transmission (skin, breast milk, fecal, vaginal, and oral), 25 mother and infant pairs were recruited and the infants longitudinally sampled from birth up to the first 4 months of life. Using high-resolution shotgun metagenomics with improved strain-level computational profiling of known and poorly characterized microbiome members, we followed the transmission and assessed the impact of the maternal microbiomes on the development of infant oral and fecal microbial communities from birth to 4 months of life.

5We enrolled 25 healthy pregnant women who vaginally delivered healthy newborns at full term. For each mother, we sampled the stool as a proxy of the lower intestine, and four additional body sites: the oral cavity (tongue dorsum swabs), the skin (intermammary cleft swabs), the vagina (vaginal introitus swabs), and the breast milk. Each newborn was sampled at two sites, the gut and the oral cavity, from birth up to 4 months postpartum (Figure 1A). All infants were exclusively breastfed at 3 days, 96% at 1 month, and 56% at 4 months. Of the 44% of non-exclusively breastfed infants at 4 months, 16% of the infants were exclusively formula-fed (Table S1A). All samples were shotgun sequenced, yielding a total of 216 high-quality metagenomes (STAR Methods), with an average of 5.73 (±7.26) Gbases per sample after quality control. DNA extraction from breast milk was not feasible in the majority of the cases, and when enough DNA was recovered a large fraction was found to be of human origin; thus, no high-quality milk microbiome samples were retained for further analysis.

6Quantitative taxonomic profiling performed with MetaPhlAn2 revealed that the diversity (Figure 1B), structure (Figure 1C), and composition (Figure S1) of the microbial communities are, as previously described, distinct at each body site. The diversity of the infant microbiome was significantly lower than all the maternal microbiomes except for the Lactobacillus-dominated vaginal microbiome (Figure 1B), which is known to have a low diversity, and was confirmed by rarefaction analysis to account for variance in sequence depths (Figure S2A). The gut and tongue microbiomes in infants were instead found to have a very high inter-subject variability, particularly at early time points (Figure 1D), compared with that of the mothers (Figure S2B). Interestingly, although the infants were all vaginally delivered, clustering based on initial sampling revealed that the early infant microbiomes did not consistently resemble one specific maternal body site (Figure 1C). For example, in some cases the early infant fecal samples (at 1 and 3 days) clustered with the maternal vaginal samples, while in other instances they clustered with the mothers' fecal samples. The high inter-subject diversity and lack of uniformity in the composition of the infant microbiomes at 1 day of life suggest that the initial exposure and seeding of the microbiome is largely stochastic, with each infant being influenced to a varying degree by the different maternal microbiomes (i.e., vagina, skin, oral, fecal), by potential prebirth in utero microbial acquisition, or by contact with other environmental sources.

7We observed a high species diversity in the infant fecal microbiomes at the first time point (T1, within 24 hr of birth; Figure 1B). This diversity decreased over the first week postpartum (p < 0.05 for both T2 and T3) before recovering over time (Figure 1B). Interestingly, the gut microbiome of exclusively breastfed infants at T4 (4 months) showed a lower diversity than the ones that gradually switched to formula feeding (Figure S2C). The high diversity at day 1 (T1) reflects the rapid influx of microbes, the pioneering microbiome, from maternal as well as other environmental sources after birth, and is consistent with a previous report using 16S rRNA taxonomic profiling. Some of these microbes are present only transiently in some infants (Table S2A) as they are probably poorly adapted or unsuited to colonize the infant lower gastrointestinal tract and, as such, are easily lost or replaced. Others are instead more likely to be true early colonizers. In particular, we observed that some species (e.g., Alistipes putredinis, Clostridium innocuum, Haemophilus parainfluenzae, Prevotella melaninogenica, and Streptococcus parasanguinis) found in the infant stool at day 1 (T1) were not present at subsequent time points (see Table S2A for the full list). This was expected as P. melaninogenica, C. innocuum, and Lactobacillus crispatus are not typically associated with the infant fecal microbiome, never being found to constitute more than 1% abundance in any of the three metagenomic datasets available for infants in curatedMetagenomicData (Table S2B). Among the species that were lost in the infants between day 1 and the subsequent time points, 80% were shared with at least one body site of the respective mother (Table S2A). Many of these shared but transient microbes most likely originated from maternal body sites other than the maternal stool (11 times from tongue dorsum, 1 from vagina, and 5 from skin), which could suggest their unsuitability to colonize the infant gut. In contrast, other more typical fecal species (e.g., Bacteroides vulgatus, Bifidobacterium longum, and Bifidobacterium breve) persisted from birth up to at least 4 months of age, when sample collection ended, suggesting that they are indeed colonizing the infant gut.

8Prepartum, the infant is subjected to largely anaerobic conditions in utero, but at birth the infant gut is mostly facultative aerobic, i.e., oxygen permissive. At day 1 (T1), we found more strict anaerobic species (on average 4.7 species per sample) than aerobic (0.4) or facultative aerobic species (4.4), albeit present at rather low abundances (16.2% in relative abundance; Figure S3). Of these strict anaerobes, 40% of the species were in common with the corresponding mother. On the third day (T2), these strict anaerobes had drastically decreased in number by 26% in the stool and 75% in the tongue dorsum. Over the same period, the abundance of facultative anaerobes increased in the stool (+9.1%) and slightly decreased in the tongue dorsum (−6.3%). These facultative anaerobes are known early colonizers of the infant gut, and these pioneering species mediate the shift from aerobic to anaerobic conditions typically associated with the adult state. Probably as a result of this transition toward anaerobic conditions, starting from the first week of age (T3), the number and the relative abundance of strictly anaerobic species increased over time (Figure S3).

9This process of early and rapid acquisition of microbial species followed by selection and succession is reflected by inter-infant microbiome distances, which, after the third day, show a decrease as the infant microbiomes converge toward a more defined composition (Figure 1D). This is explained by the decreasing effect of the original direct seeding of the infant microbiome from different maternal body sites in different mother-infant pairs. Nevertheless, the infant gut microbiome at 4 months is still markedly different from that of the mother (Permanova on Bray-Curtis dissimilarity, p < 0.0001), confirming that a longer time window is required to fully appreciate the maturation process from infant to an adult-like state.

10The infant gut microbiomes displayed a large proportion of species in common with their mothers for all mother-infant pairs. At day 1 (T1), about 50% of the microbial population in the infant gut belonged to species also present in at least one of the sampled maternal body sites (Figure 2A), and this fraction was relatively stable over time (50.7% at day 1 [T1], 48.3% at day 3 [T2], 52.2% at 1 week [T3], 37% at 1 month [T4], and 61.2% at 4 months [T5]). Shared species were notably in lower relative abundance in the mothers compared with their infants (between 2.7% and 5.5% in the mothers, depending on the reference infant time point). This suggests that it is the fitness of the microbial organisms reaching the infant gut that plays a greater role than the quantitative contribution of microbial seeding occurring for each species. The species observed in the infants but not shared with their respective mothers are likely to have been acquired from environmental sources, including other individuals that have had contact with the infant. In the oral cavity, the relative abundance of shared species between infants and any maternal microbiomes was even more pronounced with 77.6% (day 1, T1) and 95.4% (day 3, T2) of the infant microbiome being shared with the mother (Figure 2B). Also in this case, the shared species were present at low relative abundances in the maternal tongue dorsum compared with the infants (5.7% at day 1 [T1] and 6.6% at day 3 [T2]).

11We then considered the species shared between the infant and their mothers, but not with the other mothers in the cohort. Infants at day 1 (T1) shared significantly more microbial species with their mothercompared with other mothers (p < 0.05; Figure 2D). This was confirmed when looking at the fraction of shared species over the total number of species (p < 0.01) and at the cumulative relative abundance of the shared species (p < 0.05; Figure 2D). At 3 days of age (T2), infants still significantly (p < 0.01) shared a larger fraction of species with their mother compared with others, before gradually losing their species-level remembrance of the maternal microbiome (Figure 2D).

12To identify the source of the infant microbiome, we then analyzed separately the four sequenced maternal microbiomes as potential reservoirs of microbial transmission. A reanalysis of the Human Microbiome Project data confirmed that the compositions of the adult microbiomes belonging to the four body sites considered here are very distinct, with no species shared at comparable abundances across body sites, and this allowed us to compute the lists of body site-specific microbes reported in Table S3. All the maternal body sites contributed to the common mother-infant species with the mothers' stool microbiome accounting for 22.1% of the overall microbial abundance in the infant gut followed by the vagina (16.3%), the oral cavity (7.2%), and the skin (5%). Over time, the abundance of typical vaginal, oral, and cutaneous species decreased, suggesting that these species are likely transient inhabitants of the lower gastrointestinal tract. For instance, most of the vaginal species, which constituted up to 16.3% of the total abundance in the infant stool at day 1 (T1), were either lost or at undetectable levels by 1 week of age (T3). Interestingly, while abundant in the vaginal community, lactobacilli were rarely identified in the infant guts (Figure 2C). This is perhaps a consequence of lactobacilli being acidophilic, preferring a pH of approximately 3.5, and the infant gut being near neutral. This might also explain the persistence of the more pH-tolerant (pH 5.5) vaginal species, Gardnerella vaginalis and Atopobium vaginae, at least over the first days of life (Figure 2C), although other explanations involving for example nutrient requirements are also plausible. The infant oral microbiome mirrors the trend observed in the stool, with the initial presence of species common to multiple maternal body sites, rapidly followed by the predominance of species more typically associated with oral microbial communities, at 3 days of life (T2; Figure 2B).

13To add support to vertical mother-to-infant transmission events, it is necessary to identify the same strain variants within the mother-infant pairs. Because considerable individualized strain-level heterogeneity has been observed in the human microbiome, finding the same strain in the mother and in the infant would give strong evidence of intra-pair transmission. We thus implemented a novel combination of metagenomic strain-profiling tools (STAR Methods), expanding on methods validated previously. By coupling single-nucleotide variant (SNV) profiling with gene-content-based profiling, we characterized all the strains with sufficient coverage and applied a conservative threshold (0.1 in the normalized phylogenetic distance; STAR Methods) to call strain identity across paired mother-infant samples and infer vertical transmission (Figure 3A). Although there is no current consensus on the definition of a microbial strain, here we adopt the operational definition outlined elsewhere. This defines strain identity based on genomic identity using a similarity threshold on the SNV rate or gene content, which is consistent with short-term intra-subject strain variation (after excluding strain replacement), while also accounting for sequencing noise and non-adaptive low-frequency variants.

14With this approach, we detected a total of 52 strains shared within mother and infant pairs (out of 317 cases of shared mother-infant species with typable strains), i.e., a 16.4% strain transmission rate. By comparison, we found 46 strains in infants that were in common with unrelated mothers out of a total of 6,319 mother-infant inter-pair shared species. Strain sharing between infants and unrelated mothers is thus a rare event, with a sharing rate of 0.73%. This rate is even lower than we previously observed for strain sharing among an adult worldwide population, which was based solely on SNV-based profiling, indicating that the combined approach of gene content and SNV strain profiling used here is more conservative. Overall intra-pair strain sharing was >22-fold higher than inter-pair sharing (Fisher test, p < 1 × 10−15), demonstrating the influence of the maternal source in shaping the infant microbiome.

15Considering the maternal sources of transmission, the gut microbiome was the largest donor of the infant-acquired strains (Figure 3B). Common strains were also found in the maternal skin and vaginal microbiomes, but to a lesser extent. The least important route of transmission appears to be the oral cavity, with little evidence of strain sharing between mother and infant. The number of strains in common between mother and infant gradually decreases over time, with 23 at T1 (1 day) and 28 at T2 (3 days), compared with 10 at T3 (1 week), 10 at T4 (1 month), and 6 at T5 (4 months). This supports the hypothesis of selection for niche-specific bacteria from the pool of maternal strains seeding the infant gut.

16Among the species for which we observed mother-infant transmission, Escherichia coli can be tracked very effectively using pangenome analysis because of its genomic plasticity and large set of accessory genes. Transmission and persistent colonization by E. coli strains were evident in three cases (Figure 3C) all belonging to the E. coli B2 phylotype, known for enhanced persistence in the infant gut. While other E. coli phylotypes were present in infants, none of these were shared with their mothers (Figure 3C), which is suggestive of a non-maternal source for these strains. Other typically gut-associated species belonging to bifidobacteria (Figure S4) and Bacteroides (Figures 3D and 3E) displayed clear maternal routes of transmission confirmed by SNV identity patterns. As a validation for inferring strains from metagenomes, a comparison was made with strains cultivated from sample aliquots for a subset of the mother-infant pairs. Metagenomically inferred strains were near identical (>99% nucleotide identity) to those identified via single isolate sequencing (Figure S5). Cultivation also confirmed that breast milk is yet another maternal source of bacterial strains colonizing the infant, with two genomes of B. longum and Bifidobacterium bifidum strains isolated from the same milk sample (STAR Methods) recovered in the corresponding infant stool (Figure S5).

17Vertical microbial transmission from the mother to the infant can either be transient or lead to longer-lasting colonization of the infant gut. Of the vertically transferred strains, 17 were identified at more than one time point in the infant (Figures 4A and S6). In 12 of these 17 cases, after the first occurrence of the maternally acquired strain we found no subsequent replacement by another conspecific strain, i.e., 70.5% of the strains were retained and 29.5% replaced. In contrast, the 163 strains present in the infant at more than one time point, but without evidence that the mother was the source, were found to be replaced in 119 cases (73% replacement) and retained in 44 (27% retention). Vertically transmitted strains therefore seem to have a better fitness for colonization than strains without evidence of acquisition from the mother (70.5% versus 27% stable colonization, Fisher test, p < 0.001). This supports the intriguing hypothesis that maternal strains are likely to be more ecologically adaptable in the infant compared with non-maternal strains.

18We next investigated the total strain heterogeneity for each species in the microbiome of the infants compared with that of the mothers. To estimate conspecies strain heterogeneity and dominance, we analyzed the number of polymorphic nucleotide positions in the single-copy marker genes of each detected strain, as well as the average frequency of the dominant allelic variant in polymorphic positions. The analysis of maternal gut samples confirmed that the adult human gut tends to harbor only one strain of a given species, with an average fraction of polymorphic sites of 0.31% (Figure 4B). The infant gut microbiome at day 1 (T1) instead has a very high conspecific strain heterogeneity with 6.1-fold more polymorphisms than the mother (p = 1 × 10−7). As observed, the early infant microbiome at day 1 postpartum is characterized by a high species diversity (Figure 1B), which is thus also accompanied by a high strain diversity, further suggesting that the pioneering microbiome is a complex community of microbes shaped by the process of ecological selection over time. Correspondingly, at later time points there is a decrease in the intra-species polymorphic rates up to 1 month (T4), to levels comparable with those of the mothers (no significant difference at 1 month compared with the mother). Simultaneously, a higher relative frequency of the dominant strain is observed (Figure 4B). Samples collected from the infant at 4 months of age (T5) then suggest that the strain diversity is increasing and remains significantly higher than the diversity in the mothers (p = 0.0014), potentially as a consequence of the increased exposure of the infant to other possible sources of microbial seeding from the environment. Comparing the conspecific strain diversity of the infant over time with the other maternal body sites (Figure 4B), we identified markedly different levels of heterogeneity (Table S4), with the maternal tongue dorsum significantly more strain diverse than the infant gut (p < 1 × 10−10 for all time points). The maternal skin and vaginal microbiomes have instead a strain diversity in line with that of the infant stool (Table S4). Interestingly, and in contrast to the stool, the maternal oral strain diversity compared with infants is significantly higher (p = 2 × 10−9 at T2, t test). Nevertheless, in the infant oral cavity, we identified the same pattern observed in the gut, namely a high species and conspecific strain diversity (Figures 1B and 4B) followed by a rapid decline in species and strain heterogeneity due to selection, which is observed to start after a few days postpartum.

19The oral cavity is the gateway to the gastrointestinal tract, but the analysis of the oral and gut microbial communities in adults has shown them to be distinct with minimal overlap. To our knowledge, the extent of species and strain sharing in the oral and gut of infants has not been previously characterized. In our cohort, we found an average of 9.8 species in common between the stool and the tongue dorsum samples in the infant at day 1 (T1) and 7.2 species at 3 days of life (T2). As a baseline of comparison, the average in the mothers was significantly lower with only 5.5 shared species (p < 0.001 for the two time points T1 and T2; Figure 5A). In both infants and mothers, the common species were more abundant on the tongue dorsum compared with the stool (Figure 5A). However, while for the mothers almost all oral-gut shared species were at very low abundance in the stool compared with the tongue dorsum (average of 0.96%), in contrast in the infant they represented about a quarter of the total abundance of the stool microbiome (average of 24.85% and 27.09% for T1 and T2, respectively). This suggests that the oral-gut axis in infants is a rather permissive interface, and that continuous seeding of the infant gut via the oral cavity is more relevant in infants than in adults, perhaps as a consequence of reduced acid production in infants at birth as the gastrointestinal tract develops. Interestingly, the number of shared oral-gut species already decreased in the infant by day 3; however, the abundance of the shared species increased in the oral cavity, suggesting that the orally acquired species at least transiently colonize the gut.

20In the infant, the most commonly shared oral-gut species at day 1 (T1) were Gardnerella vaginalis, Propionibacterium acnes, Prevotella bivia, Atopobium vaginae, and Prevotella melaninogenica, while at day 3 (T2) they were mainly Rothia mucilaginosa, Streptococcus parasanguinis, and Streptococcus salivarius (Tables S5A–S5D). When looking at strain identity within the shared oral-gut species, we found that S. salivarius (Figure 5C) and R. mucilaginosa (Figure 5B) were the species with the highest number of shared strains (Table S5E). These results suggest that S. salivarius and R. mucilaginosa might have an increased capacity to survive in both the oral cavity and the gut, at least for a limited time.

21To perform strain profiling for microbes belonging to poorly characterized species without available genomes, we expanded our analysis by performing metagenomic assembly on each sample followed by contig binning, phylogenetic profiling, and whole-genome strain identity inference (STAR Methods). Overall, we reconstructed 1,132 metagenome-assembled genomes (on average five per sample; Table S6A) with sufficient quality (≥50% completeness, ≤5% contamination) to be amenable for strain tracking. Of these, 763 genomes could be assigned to a known species by applying a 95% percent identity threshold on the whole sequence (STAR Methods) and were therefore not further processed because these strains were captured by the reference-based methods already considered above. However, the remaining 369 genomes (Figure 6; Table S6A) did not belong to any of the 13,575 microbial species for which at least one reference genome is available (STAR Methods), including 36 that could not even be assigned below the level of family. The genera containing most of the unknown species were Streptococcus (32 genomes), Clostridium (31), and Prevotella (31).

22Next, we compared the 369 taxonomically uncharacterized genomes against each other to identify the presence of the same strain in different metagenomic samples. Using a strict threshold of 99.5% identity over the full length of the genomes, we identified eight vertical transmission events (Figure 6; Table S6B). In six cases the strain sharing was between the mother and infant gut microbiomes. Two of these strains belonged to uncharacterized species in the Akkermansia and Bacteroides genera (less than 89% identity with the closest available genomes over less than 75% of the length), while for the other four strains classification was even more challenging and we could only infer that they belonged to four different phyla (Verrucomicrobia, Proteobacteria, Bacteroidetes, and Firmicutes; Table S6B). In addition to the fecal transmission routes, two vertical transmission events were also observed from other body sites with an uncharacterized Clostridiales strain (99.9% of similarity) shared by the maternal vaginal community and the stool of the infant, and an unknown Leptospira strain (99.9% of similarity) shared by the skin microbiome of the mother and the saliva microbiome of the infant. There was only one case of a strain from an unknown species with 99.9% similarity within an unrelated mother-infant pair, strongly confirming the occurrence of vertical transmission for the eight genomes above (Fisher test, p < 1 × 10−9) and confirming that uncharacterized species have a role in the mother-to-infant microbial seeding.

23We investigated here the early acquisition and development of the infant gut and oral microbiomes, and in particular the role of different maternal sources in this process, by means of a longitudinal multiple body site metagenomic approach. Among our main findings is the very high microbial diversity and strain heterogeneity in the pioneering infant gut microbiome even in the first day of life, which dramatically decreases within the first week before recovering and gradually increasing over the next 4 months. While we cannot discount the possibility of intra-uterine microbial acquisition, this suggests early seeding with an overall species diversity and strain heterogeneity far higher than previously appreciated, followed by steep selection forces that maintain only part of this early biodiversity. The selection process is corroborated by the decreasing strain heterogeneity in the developing infant microbiome and the initial increase of facultative anaerobes, which are subsequently replaced by strict anaerobes consistently with the biochemical changes of the infant gut environment. Far from being a static process, microbial seeding from maternal sources is continuous, with some species and strains appearing in the infant at later time points. We thus describe how the microbial colonization process in the infant reflects a balance between influx of microbial strains and a process of niche selection. This balance is likely the key for the physiological development of the infant microbiome and should be further studied to unravel potential links with pathologies in childhood.

24Another key finding of our study is that we identify microbial strains present in the infants for which there is strong evidence of transmission from their mothers, and that these strains are more likely to adapt to and persist in the infant gut than non-maternally acquired strains. This reinforces the importance of this vertical mother-to-infant microbial transmission from multiple sources, because even though the maternal microbiomes cannot explain many microbial strains present in the infant, the transmitted strains appear crucial in the developing microbiome. The mechanisms of this phenomenon should be further investigated and may be related to a combination of prebirth transmission, shared environmental factors, and common genetic factors that could partially explain the mother-infant strain specificity. Nevertheless, if we consider vertical microbial transmission as a physiological process under evolutionary pressure in recent human history, the study of vertically transmitted strains can provide the basis for better understanding the impact of non-vaginal birth (C-section) and non-exclusive breastfeeding.

25The methodological approaches in this work are a novel combination of reference-based and assembly-based computational profiling that enables us to comprehensively describe mother-to-infant strain transmission and strain-level dynamics in the infant microbiome. The adoption of computational profiling tools with a resolution at the level of individual strains is key because no direct evidence of microbial transmission can be inferred when bacteria are categorized solely at the species level. In this work, we exploit and model the very high inter-subject variability of strains in the general population, and suggest transmission only for the cases in which a mother and her infant have a strain similarity substantially higher than that found between unrelated subjects. Overall, we find 62 cases with strong evidence of strain transmission, eight of which are from species that are currently uncharacterized. We also find evidence of strains originating from multiple maternal sources, with the vaginal, skin, oral, and gut communities all contributing to the early infant microbiome. However, even after a few days postpartum, the contribution of the vaginal and skin microbiome already decreases. Here we focus on the mother, perhaps the most important familiar relationship in the development of the infant microbiome. However, the same techniques can be applied to describe other routes of transmission, including family members other than the mother, and the hospital rooms. These sets of microbial sources have not been studied together, and their integrated analysis will further contribute to understanding the mechanisms of early microbiome acquisition and subsequent development.

26Further information and requests for resources, reagents, and software should be directed to and will be fulfilled by the Lead Contact, Nicola Segata (nicola.segata@unitn.it).

27A total of 25 pregnant women (26-43 years of age, BMI before pregnancy between 16.5 and 29.4) were recruited by Santa Chiara Hospital in Trento, Italy, between April 2015 and July 2016. The protocol of this study was approved by the Ethics Committee of Santa Chiara Hospital (Trento, Italy) and the Ethics Committee of the University of Trento. Parents were asked to provide informed consent and complete questionnaires regarding pregnancy data, maternal diet, medical records, and lifestyle before and during the pregnancy (Table S1A).

28Exclusion criteria included delivery by Caesarean section, pre-term birth (<37 weeks), body temperature >38°C, birth weight <2.5 Kg, antibiotic treatments during pregnancy, intention to avoid formula feeding at least for the first six months. The infant feeding practices as well as antibiotic usage and complete medical records across the sampling period were also collected (Table S1A). At 4 months of age, 56% of infants were exclusively breastfed (14 out of 25), while 12% (3 out of 25) of mothers reported mixed feeding and 16% exclusive formula feeding (4 out of 25). For four pairs, the sampling was interrupted before the age of 4 months, and therefore no metadata in terms of feeding are available.

29The samples collection procedure was based on HMP sampling guidelines. Skin samples were collected using Catch-All-Swabs (Epicentre Technologies, Wisconsin, USA) shortly after birth but before the skin-to-skin contact with the infant, by swabbing the upper area of the maternal breast (intermammary cleft). After pre-moistening with 2 ml SCF-1 buffer (50 mM Tris buffer, pH 7.6, 1mM EDTA, pH 8.0, and 0.5% Tween-20) contained in a 15 ml sterile screw top collection tube (Sarstedt, Nümbrecht, Germany), the swab head was rubbed back and forth for approximately 30 seconds over the area (repeating twice) before the swab was returned to the buffered solution. Vaginal swabs were collected before delivery, as soon as possible upon arrival at the delivery ward to reduce potential blood contamination. The swab was rubbed 5 times, with a circular motion, in the vaginal introitus and then the swab head was placed in a 15 ml sterile screw top collection tube containing 2 ml SCF-1 buffer. Maternal tongue dorsum swabs were collected shortly before or during delivery by rubbing a swab on the central area of the back of the tongue for approximately 5 seconds. The swab head was then placed in a 15 ml collection tube containing 2 ml SCF-1 buffer. The same procedure was used for sampling the infant tongue dorsum, at time of the collection of the first stool sample. Maternal breast milk was self-collected by the mothers starting at one day after delivery, using gloves to avoid skin contamination. Stool samples from the mother were collected during or shortly after the delivery by the hospital staff, using collection tubes specific for faecal material (Sarstedt, Nümbrecht, Germany). All infant stool samples were self-collected by the mother, following a detailed protocol. The samples collected directly at the hospital were frozen at -20°C immediately after the collection and moved to a -80°C facility within a week, where they remained stored until further analysis. Additional aliquots were collected and stored in 20% glycerol and kept at -20°C for the cultivation experiments. After leaving the hospital, normally three days after delivery, the mothers performed the collection of the infants’ stool samples at home and put the samples immediately at -20°C, which were then delivered to the hospital staff within 12 hours.

30DNA was extracted using the PowerSoil DNA Isolation Kit (MoBio Laboratories Carlsbad, USA), as described in the HMP protocol. For the stool samples, a preliminary heating step (65°C for 10 minutes, 95°C for 10 minutes) was performed before extraction. For samples collected via swabbing (skin, oral, and vaginal) the head of the Catch-All Sample Collection Swab was removed and put in the PowerBead Tube. The 2 ml SCF-1 specimen-containing buffer was centrifuged at 1000g for 5 min and added to the PowerBead Tube for cell lysing performed using the MOBIO Vortex Adapter (MO BIO Catalog No. 13000-V1). DNA was recovered in 10 mM Tris pH 7.4 and quantified using Qubit 2.0 (Thermo Fisher Scientific, Massachusetts, USA) fluorometer as per manufacturer's instructions. Sequencing libraries were prepared using the NexteraXT DNA Library Preparation Kit (Illumina, California, USA), following the manufacturer's guidelines. The sequencing was performed on the HiSeq2500 (Illumina, California, USA).

31Out of the 225 collected samples subjected to sequencing, 216 provided more than 5.1 x 10ˆ9 total reads (average 23.96 x 10ˆ6 reads per sample) and were therefore used for the downstream analysis. Of these samples, 119 (55%) were stool samples (21 from the mothers and 98 from the infants), 15 (6.9%) were skin swabs from the mother, 63 (29.2%) were oral cavity swabs (24 from the mothers and 39 from the infants), and 19 were derived from vaginal swabs (8.8%).

32The generated raw metagenomes were processed with FastqMcf by trimming positions with quality <15, removing low-quality reads (mean quality <25), and discarding reads shorter than 90 nt. Human and bacteriophage phiX174 (Illumina spike-ins) DNA were then removed using BowTie2 to map the reads against the reference genomes. The adapters were also discarded by trim_galore (http://www.bioinformatics.babraham.ac.uk/projects/trim_galore/) with parameters "-q 0 --nextera --stringency 5". All samples providing less than 50,000 reads were excluded from the downstream analysis. The number of reads for each sample after preprocessing is reported in Table S1B.

33Species-level quantitative taxonomic profiling was performed using MetaPhlAn2 (version 2.6) on the post-processed reads. MetaPhlAn2 estimates the relative abundances of each known microbial species. These relative abundances are proportional to the underlying absolute species concentrations that cannot be quantified from metagenomic data alone. Taxonomic profiles included bacteria, archaea, microbial Eukaryotes, and viruses, and were inferred by MetaPhlAn2 using the ∼1 M unique clade-specific marker genes identified from ∼17,000 reference genomes (∼13,500 bacterial and archaeal, ∼3,500 viral, and ∼110 eukaryotic). Rarefaction analysis was performed using Seqtk (https://github.com/lh3/seqtk) at 5 million reads per sample; 57 samples (28 of which from infants) were excluded because of insufficient sequence depth (less than 5 million reads per sample after sub-sampling).

34Strain-level analysis was performed using a combination of gene-content-based profiling using PanPhlAn, and single-nucleotide variant profiling using StrainPhlAn. PanPhlAn was applied on the preprocessed metagenomes using default parameters generating a presence/absence gene-family profile independently for each sample and each species present in the reference database. A species-specific gene-family matrix was obtained by combining profiles from different samples and was further processed with RAxML to generate a species-specific phylogenetic tree. Comparison among different trees was performed by normalizing distances in each tree by its median value. Finally, strain distance for any pair of samples was defined as the normalized phylogenetic distance on the corresponding tree. Similarly, StrainPhlAn was run on the preprocessed reads with default parameters and adding the options "-alignment_program mafft" and "--relaxed_parameters3″ for each reference species and by considering pair-specific markers. This generated a maximum of twenty-five (i.e., the number mother-infant pairs) phylogenetic trees for each species. A single species-specific value for each pair of samples was obtained by averaging their distances along the different trees. Strain distance was defined as for PanPhlAn including normalization of each tree by its median value. The final strain distance for any sample pair was obtained by taking the minimum value between the two strain distance values generated by PanPhlAn and StrainPhlAn. We considered a threshold of 0.1 on the strain distance for defining a pair of strains as the same strain. This threshold was chosen as the value dividing the two peaks in the bimodal distribution of all-versus-all normalized species-specific strains distances representing identical and thus possible transmitted strains (values smaller than 0.1) and clearly distinct strains (values bigger than 0.1). The same approach was adopted for all species under the assumption that this threshold is very stringent and that almost identical genomes represent almost identical strains irrespective of the species they belong to. Results reported in Figures 3A and 3B were obtained by considering the minimum strain distance value for each pair of subjects to avoid skewing of the statistics due to effects of multiple timepoints and multiple body sites effects. Strain-level phylogenies presented in Figures 3D and 3E and in Figures 5B and 5C were obtained from the StrainPhlAn analysis using RAxML (parameters: -m GTRCAT –p 1234). For the four phylogenies in the two figures we also reported the proportion of phylogenetic distance in percentage computed with respect to the total branch length of the respective phylogeny.

35Results reported in Figure 1C were computed using the Bray-Curtis dissimilarity and the MDS algorithm, both implemented in the scikit-learn Python package (scikit-learn).

36All samples were processed independently for de novo metagenomic assembly through metaSPAdes (version 3.10.1 using default parameters), discarding contigs shorter than 1000 nt. Reads were mapped to contigs using Bowtie2 (version 2.2.9; option “--very-sensitive-local”), and the mapping output was used for contig binning through MetaBAT2 (version 2.12.1; option “-m 1500”). CheckM (version 1.0.7; lineage specific workflow) was applied to the resulting bins, and only those with sufficient quality (≥50% completeness, ≤5% contamination) were considered for further analysis. We used BLASTn (version 2.6.0; default parameters) to map the contigs of each reconstructed bin against all the microbial reference genomes (including bacteria, archaea, viruses, and microeukaryotes) available in the NCBI repository as of September 2016 for a total of 13,575 unique and named species. Each position of a given contig was assigned to the hit with the highest bitscore, and then the median percent identity across the mapping positions was computed. The average percentage identity for a genome was computed by averaging the percent identity values amongst its contigs. For detecting potential mother-to-infant transmissions, we considered only bins with an average percent identity below 95%, as genomes with available close references were already considered by the assembly-free approach described above. The 369 genomes without a close reference were then compared against each other to identify the presence of the same strain in different samples. We computed the average nucleotide identity (ANI) using the pyani tool (v.0.2.6; option “-m ANIb”) and considered a threshold of 99.5% identity over the full length of the genomes to define strain identity. Because of the use of the whole genome instead of species-specific markers and pangenomes, the strain identity thresholds for metagenomically assembled strains and StrainPhlAn/PanPhlAn profiled strains are not directly comparable. Phylogenetic placement of the reconstructed genomes was obtained by running PhyloPhlAn2, and visualized through GraPhlAn.

37Bifidobacterium strains were inoculated in Man-Rogosa-Sharp (MRS) (Scharlau Chemie) supplemented with 0.05% (wt/vol) L-cysteine hydrochloride and incubated in an anaerobic atmosphere [2.99% (vol/vol) H2, 17.01% (vol/vol) CO2, and 80% (vol/vol) N2] in a chamber (Concept 400, Ruskin) at 37°C for 16 h. Chromosomal DNA was extracted as described previously. Library preparation and Illumina sequencing were performed at the GenProbio srl (Parma, Italy) and previously described.

38We validated the microbial transmission results for bifidobacteria by searching for the presence of the 12 assembled genomes isolated in the metagenomes. We first mapped the metagenomics reads against the 12 isolates genomes using Bowtie2 (parameters: --very-sensitive-local –k 100000 –no-unal). Then, we performed a SNV-rate by analyzing the “.sam” output from Bowtie2 through Samtools and custom Python scripts. For each isolate genome mapped against a single metagenome, we considered the sites with at least two high-quality reads covering the single position in either forward or reverse orientation, as defined in the “DP4” Variant Call Format field.

39The permanova analysis has been performed using the scikit-bio Python package (scikit-bio). When not specified otherwise, t-test was used to calculate p-values. All the other statistical analyses have been performed with open source software mentioned and referenced in the description of the analyses. The number and description of the samples are reported in Table S1B, and the statistical analyses always refer to the whole set of samples in the specific condition of interest.

40All metagenomes were deposited and are available at the NCBI Sequence Read Archive (SRA) under BioProject number PRJNA352475 and SRA accession number SRP100409. Taxonomic profiles along with metadata are available in the curated MetagenomicData package. Assembled contigs are available at https://www.dropbox.com/s/7pygii8khtrriw0/genome_bins.tar.bz2?dl=0. All the software packages used in this study are open source and publicly available and the new software developed for this study is contained in the new releases of PanPhlAn and StrainPhlAn available at https://bitbucket.org/CibioCM/panphlan/wiki/Home and https://bitbucket.org/biobakery/biobakery/wiki/strainphlan respectively.