The skin microbiome in the first year of life and its association with atopic dermatitis

Early‐life microbial colonization of the skin may modulate the immune system and impact the development of atopic dermatitis (AD) and allergic diseases later in life. To address this question, we assessed the association between the skin microbiome and AD, skin barrier integrity and allergic diseases in the first year of life. We further explored the evolution of the skin microbiome with age and its possible determinants, including delivery mode.

population-based birth cohort study, representing 970 samples. Clinical investigations included skin examination and skin barrier function measured as trans-epidermal water loss (TEWL) at the site and time of microbiome sampling at 3, 6, and 12 months.
Parental background information was recorded in electronic questionnaires, and delivery mode (including vaginal delivery (VD), VD in water, elective caesarean section (CS) and emergency CS) was obtained from maternal hospital charts.
Results: Strong temporal variations in skin bacterial community composition were found in the first year of life, with distinct patterns associated with different ages.
Confirming our hypothesis, skin bacterial community composition in the first year of life was associated with skin barrier integrity and later onsets of AD. Delivery mode had a strong impact on the microbiome composition at birth, with each mode leading to distinct patterns of colonization. Other possible determinants of the skin microbiome were identified, including environmental and parental factors as well as breastfeeding.
Conclusion: Skin microbiome composition during infancy is defined by age, transiently influenced by delivery mode as well as environmental, parental factors and breastfeeding. The microbiome is also associated with skin barrier integrity and the onset of AD.

K E Y W O R D S
atopic dermatitis, cohort study, infancy, microbiome, skin

G R A P H I C A L A B S T R A C T
The skin bacterial community composition varies between individuals after birth, then progressively converges during the first year of life. The skin microbiome composition and diversity at birth is associated with later onset of atopic dermatitis with associations found between skin barrier function, dry skin, and bacterial communities. Delivery mode and location are major determinants of the skin microbiome at birth. Abbreviations: NMDS, non-metric multidimensional scaling; PreventADALL, Preventing Atopic Dermatitis and ALLergies in Children; TEWL, trans-epidermal water loss

| INTRODUC TI ON
Atopic dermatitis (AD), which affects one fifth of the world population and often starts in infancy, is characterized by a dysfunctional skin barrier, inflammation, and a dysbiotic microbiome on lesional skin. 1 Skin barrier impairments may precede AD development, suggesting that preventive strategies could be implemented based on early risk identification. 2,3 Recent advances have revealed a strong relationship between early-life microbial colonization and the development of the immune system, 4,5 and the associations between the gut and skin microbiome and allergic diseases such as asthma and AD have been extensively studied. 4,6,7 Although the skin microbiome development is influenced by many factors including genetics, antibiotics, diet, and environmental exposures, maternal transfer during delivery is the most obvious source of a newborn skin microbial colonization. 8 While studies have addressed the impact of delivery mode, in particular caesarean section (CS), on the microbiome, few distinguished between elective and acute/emergency CS and the possible role of various delivery modes on the development of allergic diseases remains unclear. [9][10][11] In AD, lesional skin is typically colonized by Staphylococcus aureus, whose abundance decreases upon treatment as skin microbial diversity is restored. [12][13][14][15][16] While S. aureus outgrowth seems to contribute to disease recurrence it remains unclear whether it is first a consequence or a cause of AD. In contrast, skin colonization by commensal Staphylococci species has been associated with lower incidence of AD in babies, suggesting these bacteria may also play a role in the disease development. 17 High transepidermal water loss (TEWL) is also characteristic of AD, reflecting an impaired skin barrier, both in lesional and non-lesional skin. 18 Few studies have investigated the relationship between TEWL and skin microbiome and to our knowledge there are no previous human studies focusing on infants.
In this context, we hypothesized that the skin microbiome in early life may be associated with the later development of allergic diseases. To verify this we explored associations between the skin bacterial communities of the lateral upper arm and variables covering atopic dermatitis and skin barrier at four timepoints in the first year of life. To gain further insight into the determinants of the skin microbiome in infancy, we also assessed associations with delivery mode, genetics, environmental and parental factors, and breastfeeding. Vaginal delivery (VD) and CS have different impact on the skin microbiome at birth, and we further hypothesize that different types of VD and CS may also differently affect it. In particular, we set out to compare normal VD to VD in water, elective CS and emergency CS, the main difference between the two later being the level of exposure to the vaginal flora.
To our knowledge this is the largest longitudinal study assessing the skin microbiome in the first year of life and the first to investigate the impact of delivery mode while considering VD in water and distinguishing between elective and emergency CS.

| ME THODS
See Appendix S1 for detailed methods.

| Study population
The present study included 346 newborns selected from the Preventing atopic dermatitis and allergies in children (PreventADALL) study, 19 a population-based randomized control trial (RCT) and observational study enrolling 2396 mother-child pairs in Oslo, Østfold (both Norway) and Stockholm, Sweden. Children were recruited antenatally and randomized at birth into four groups: observation only (no intervention), food intervention (early introduction of peanut, milk, wheat, and egg from 3 to 6 months), skin intervention (oil bath and facial cream from 2 weeks to 9 months), or both food and skin interventions. 20,21 The primary outcomes of the study was the prevention of food allergy at 3 years and the prevention of AD at 1 year. The present nested study includes the first participants from whom clinical data and skin microbiome samples were available longitudinally. The study size was estimated to include at least 50 AD patients based on an anticipated prevalence of 23%. 22 Although only participants born in Norway were included in the nested cohort, its characteristics were similar to the ones of the larger RCT (Table S1).
Informed written consent was obtained from all pregnant mothers upon inclusion, and again from both parents upon inclusion of the child. The PreventADALL study has been approved by the Regional Ethical Committee for Medical and Health Research Ethics in South-Eastern Norway (2014/518) as well as in Sweden (2014/2242-31/4) by the Regional Ethical Trial Committee of Stockholm. It is an ongoing study that has been approved for data collection until 2044.

| Skin examination
Health personnel examined the skin by visual inspection and palpation as previously described. 23 Eleven predefined skin areas were rated in terms of no, mild, moderate, or severe dry skin. 3 Atopic dermatitis was defined using either the diagnostic criteria of the UK Working Party 24 and/or those of Hanifin and Rajka 25 during examinations at 3, 6, and 12 months. 26 Eczema was defined as the presence of eczematous lesions observed by a medical doctor, with the exclusion of differential diagnoses to AD.

| Bacterial 16S rRNA gene amplicon sequencing
Skin bacterial communities were assessed by high-throughput sequencing of the V4 hyper-variable region of the bacterial 16S rRNA gene as previously described, 27
2.6.1 | Amplicon sequences processing and quality control Bacterial 16S rRNA gene amplicon sequences were processed into a table of amplicon sequence variants (ASV) using the dada2 pipeline (R package dada2 version 1.18.0). 28 Samples with fewer than 20,000 reads were excluded from the dataset ( Figure S1A) and ASVs below 1% prevalence or unassigned at Phylum level were filtered out. Reads were rarefied at 20,000 reads per sample using random sampling without replacement (hereafter referred to as the rarefied ASV counts) and a Hellinger transformation was applied (hereafter referred to as the transformed ASV counts). Rarefaction analysis showed a sufficient sequencing depth to detect most taxa ( Figure S1B). Eleven samples collected at 6 and 12 months and displaying a diversity lower than the minimum observed at day 1 were also removed ( Figure S1C,D). These samples were not associated with any sign of disease, based on the collected clinical data (Table S2). Burkholderiaceae was identified as a potential contaminant based on 11 negative PCR controls that led to read assignments (encompassing 717,721 reads assigned to 300 ASVs, 231 of which found in less than three samples and only eight of which found in more than five samples) ( Figure S2). Given the relatively high proportion of this taxa in samples collected at day 1, it was not ignored when comparing bacterial communities between ages yet was ignored, together with ASVs assigned to chloroplasts and mitochondria, in all associative analyses.

| Statistical analysis and visualization
All boxplots represent first and third quartiles, with the median as a middle line and whiskers at the last value within a 1.5×IQR distance respectively from the upper or lower quartile, where IQR is the interquartile range. Lines in all violin plots represent the first, second, and third quartiles.

Evolution of the skin microbiome in the first year of life
The skin bacterial community composition was compared across ages using perMANOVA accounting for repeated sampling, on complete longitudinal sample sets (n = 124). Non-metric multidimensional scaling (NMDS) analysis was applied on the transformed ASV counts. Shannon diversity index (SDI) and richness were assessed based on rarefied ASV counts. A random forest (RF) classifier was trained to predict the age and extract associated features based on transformed ASV counts (R package Boruta).

Associations between skin microbiome and investigation variables
Associations between the skin bacterial community composition and categorical investigation variable was assessed separately at each age based on imputed values using perMANOVA in an additive model accounting for interactions with delivery mode and RCT interventions (day 1: n = 225, 3 months: n = 273, 6 months: n = 242, 12 months: n = 230). For each variable showing a significant association with the skin bacterial community composition (p < .05), associated ASVs were identified in a two-step process including a first feature selection based on truncated the Kruskal-Wallis test for zero-inflated data 29 (p < .01) and second a sparse partial least squares discriminant analysis (SPLS-DA) (R package spls). 30 Associations between ASVs and TEWL was assessed separately at each age using sparse partial least squares (SPLS) regression (R package spls). Associations between food allergy and the skin bacterial community composition and diversity was assessed separately.

Associations between skin microbiome and delivery mode
Constrained correspondence analysis (CCA) based on transformed ASV counts and constrained along age was applied and its first component was used to compare samples collected from babies born through different delivery modes. SDI and richness based on rarefied ASV counts, as well as TEWL were compared between delivery modes. Correlations between SDI at day 1 and TEWL were assessed within samples from the same delivery modes using the Kendall rank coefficient. and Figure S3B; Table 1 and Table S3). Infants were born at a mean (SD, min-max) gestational age of 39.7 (1.5, 35.2-42.9) with a mean (SD, min-max) birth weight of 3.6 (0.5, 2.0-5.6) kg and 45% were females. PCR controls did not pass the quality control, yet allowed identification of Burkholderiaceae as the main potential contaminant ( Figure S2). In total, 2730 ASVs were observed, encompassing 21 bacterial phyla with Proteobacteria, Firmicutes, Actinobacteria, and Bacteroidetes being the most abundant (Figure 2A,B).

| Microbiome and investigation variables
Investigation variables included 30 categorical variables with at least 10 individuals per group, plus TEWL measurement collected at same location and time as microbiome samples ( Figure S4 and Table 1 and Tables S4-S6). Individuals were grouped into two clusters globally reflecting impaired or intact skin barrier ( Figure S5A).
The first cluster was characterized by higher frequencies of eczema and AD as well as higher TEWL at 3 and 6 months ( Figure S5B,C).

| Evolution during the first year of life
Age influenced the composition and diversity of skin bacterial communities in the first year of life (Table S7). In particular, the relative   substantial differences ( Figure 2E and Figure S7). For example, Lactobacillus iners were most abundant at day 1 and dropped afterward. Enhydrobacter aerosaccus were in low abundance at day 1 and increased thereafter, while certain Streptococcus and Acinetobacter gradually increased from birth to 6 and 12 months, respectively. In comparison, Bifidobacterium and Veillonella showed transient peaks at 3 and 6 months, respectively.

TA B L E 1 (Continued)
and Table 2 and Table S8). Among these associations, nine reflected variations in skin bacterial diversity, including birth location, dry skin, AD as well as maternal AD, and food allergy ( Figure 3B and Figure S8A). Seven variables were found associated with sets of 10-187 ASVs at day 1 and 3, 6 or 12 months ( Table 2 and Table S9): Birth location was associated with 86, 59, and 22 taxa at day 1, 3 months, and 6 months, respectively, including a same Betaproteobacteriales at both day 1 and 3 months ( Figure 3C and Figure S9). The presence of dogs during pregnancy was associated with 10 and 11 ASVs at 3 and 6 months, respectively. This included higher abundances and frequencies of Betaproteobacteriales and Ralstonia at 3 months ( Figure 3C and Figure S10) both timepoints ( Figure 3C and Figure S11). Dry skin observed at 3 months was associated with 24, 59, and 23 ASVs at day 1, 3 months, and 12 months, respectively, including higher abundances and frequencies of Streptococcus at both day 1 and 3 months ( Figure S12).
Maternal food allergies were associated with 121 and 187 ASVs at day 1 and 6 months, respectively, including higher representations of Prevotella and Alloprevotella at day 1 and 6 months, respectively ( Figure S13). Measurement of TEWL at 6 months was associated with 15 and 21 ASVs at day 1 and 3 months, respectively, including Ralstonia taxa which representations correlated with higher TEWL ( Figure S14). Among the variables associated with the skin bacterial community, five were also associated with TEWL: birth location, the presence of dogs during pregnancy, dry skin on limbs at 3 months of age, AD, and maternal food allergy ( Table 2 and Figure S8C). SPLS regression did not identify any strong association between ASV abundance and TEWL ( Figure S15). Seven associations between microbiome composition and investigation variables that involved interactions between either delivery mode or RCT intervention were observed (Table S8). Food allergy was found associated with the skin bacteria community composition, as well as with higher TEWL, at 3 months (Table S11 and Figure S19).

| Comparisons between delivery modes
Differences in skin bacterial community composition were observed in the first day of life between infants born either through VD, VD in water, emergency CS or elective CS (p < .05). A significant association was also observed between delivery mode and the skin bacterial community at 6 months ( Table 2 and Table S8). The skin bacterial community of babies born by CS and VD in water were, on day 1, more similar to communities found at 3 months, as estimated by age-constrained CCA ( Figure 4A,B and Figure S16A). At day 1, lower SDI was found in babies born through emergency and elective CS compared to VD and VD in water ( Figure 4C). One hundred eighty seven ASVs were associated with different delivery modes at day 1 ( Table 2 and Table S9), including Lactobacillus, Staphylococcus, and Pseudomonas found underrepresented in elective CS as well as a Streptococcus found in higher abundance in elective CS and VD in water compared to normal VD ( Figure S16C). Inter-individual variations were observed in the pattern of colonization at this age, and no consistent, global combinations of taxa associated with specific groups were observed ( Figure S17).  (Table S8), SDI, and Richness: the skin bacterial diversity in terms of Shannon diversity index (SDI) and richness ( Figure S8A), TEWL: trans-epidermal water loss (TEWL) (Figure S8C), Ass. taxa: number of skin bacterial taxa associated with the investigatory variable, identified based on truncated the Kruskal-Wallis test and sparse partial least squares discriminant analysis (SPLS-DA) (Table S9) ( Figure 4D). In emergency CS, the skin bacterial diversity at birth was correlated with TEWL values observed at 3 months -lower SDI being associated with higher TEWL, and thus weaker skin barrier function ( Figure 4E).
Caesarean section, and in particular elective CS, was associated with a higher frequency of eczema in the first year of life. A similar trend was observed for AD and the individuals cluster reflecting weaker skin barrier, although this trend was lost when considering AD diagnosed by 3 years of age. Dry skin was observed in all infants born through elective CS ( Figure S18 and Table S10).

| The skin microbiome evolves in the first year of life
In line with previous studies, we show that the skin microbiome of

| The skin microbiome is associated with AD and skin barrier integrity
Variations in the skin bacterial community in early life were associated with later manifestations of AD, dry skin, and reduced skin barrier integrity ( Figure 5A and Table 2) as well as food allergy (Table S11), suggesting that the skin microbiome in early life may play an important role for the later development of allergic disease.
Given that (1) the skin barrier is weaker at birth and strengthen with age and (2) the skin at birth constitutes an empty niche which can be seeded by distinct sets of microbes, we hypothesize that the skin of newborn, and their immune system, is likely to be significantly impacted by microbial colonization. Large differences between individuals may therefore be expected, perhaps leading some to get primed for the later development of AD ( Figure 5B). Our results suggest that the freshly seeded newborn skin constitutes an ecosystem that converges toward stability and reduced inter-individual variations, which altogether may explain why the immature microbiome of newborns is more easily associated with AD than the more mature one of 1 year old babies. The role of the microbiome at other sites, like the gut, as well as the interplay with the immune system and metabolism, remain to be investigated.
Our finding of associations between bacterial community composition and dry skin at any location and specifically on the limb extensors is novel. The observation of a higher skin bacterial diversity in newborns and 3-month-old infants with dry skin contrasts with a previous report where emollients increased the skin microbial diversity in AD patients, who are prone to dry skin. 34 However, most of the participants presenting dry skin in the present study did not develop AD before 12 months. The association between skin bacterial communities and diagnosed AD in the first year of life is in line with previous findings from a smaller Polish infant study 35 and studies including older children and adults. 12,[36][37][38] No specific bacterial taxa was found associated with AD, which is supported by previous findings in the Polish study, 35

| The skin microbiome is associated with environmental and parental factors and breastfeeding
Associations involving environmental or parental factors, including birth location, pets, and parental allergic diseases were observed from birth to 6 months. Infants born at different locations showed strong differences in skin bacterial populations from day 1 to 6 months, which we hypothesize may be due to differences in the bacteria found in the hospital environment.
The finding that the presence of dogs during pregnancy was  protective effect against atopic dermatitis from pets and farm animals during pregnancy suggests it may be mediated by the stimulation of the newborn innate immunity. 3,40 The associations found between breastfeeding until 6 months and certain taxa at 3 and 6 months are in line with a previous smaller study and may reflect either the direct contact of the mother's milk on the skin, or an indirect mechanism involving the infant metabolism and immunity. 41 Interestingly, this association seems transient and limited to the typical period of breastfeeding, as weaning often occurs before 12 months.

| Delivery mode shapes the skin microbiome at birth
In line with previous findings, we confirm that delivery mode impacts on the skin bacterial community at birth. [42][43][44] In particular, vaginally delivered newborns harbored a higher proportion of bacteria belonging to a typical vaginal microbiome (e.g., Lactobacilli) and had higher bacterial diversity compared to CS delivered babies.  45 We observed a lower Lactobacillus abundance in elective CSs compared to emergency CSs, in line with our previous reporting of labor-induced exposure to vaginal flora during emergency CS in the PreventADALL cohort. 46 Interestingly, in the first day of life, the skin bacterial community of babies born through CS or VD in water resembled more a skin microbiome of a 3-month old baby compared to that of vaginally-born babies, whose skin bacterial composition was more clearly distinct from the one of older babies.
Overall, this highlights that the skin of vaginally-delivered newborns is mainly seeded by the vaginal flora of the mother while the skin of CS-delivered newborns is seeded by more typical skin bacteria, likely through contact with the mother's skin. We hypothesize that VD in water may have a diluting effect on the bacteria acquired in the birth canal and that a larger contact with fecal microbiota may occasionally occur through contamination of the water. Therefore, babies born this way may exhibit a pattern of bacterial colonization where the place of the vaginal flora is less prominent. Altogether, these findings highlight that all four modes of delivery are associated with distinct patterns of microbial colonization at birth.
To our knowledge, for the first time, we show that the delivery mode impacted on the newborn's skin microbiome and that variations of the latter correlated with variations of the skin barrier function ( Figure 4E). However, it is not clear whether the skin microbiome was impacting the skin barrier function, or whether a common causal factor exists which increases the risk of CS delivery while also affecting the newborn's skin barrier, independently of the skin microbiome. This may involve parental factors such as metabolic diseases and anxiety as well as genetic, environmental, and life-style factors that may affect both mothers and children. [47][48][49]

| No impact of skin intervention nor genetics were detected in the first year of life
No associations between the RCT interventions and skin bacterial composition was found, which contrasts with other studies reporting higher bacterial diversity in skin treated with emollients. 34,50,51 Surprisingly, no association between filaggrin functional deficiency and the skin microbiome were detected, despite it being responsible for impaired skin barrier and a major risk factor for AD. 52,53 This suggests that while genetics can shape the skin microenvironment, its importance in the selection of the colonizing bacteria may be out-competed by other factors, such as exposure to different sets of microbes from the mother's flora and the environment.
Overall, our study provides a strong motivation to further investigate the role of distinct skin microbiomes and the role of delivery mode in the development of allergic diseases.

| Strengths and limitations
The strengths of the present study include its longitudinal and pro- instance, one could hypothesize that only a subcategory of AD is mainly driven by the colonization of bacteria. In such a case, the AD diagnosis methods used in the present study would fail to distinguish between a more bacterial-driven AD and the other, microbiomeindependent cases. Therefore, studies including larger groups of AD patients may be needed to represent the heterogeneity of the disease and discover microbiome patterns associated with its distinct sub-categories.

| CON CLUS ION
Using a cohort including over 340 infants, we show that the skin bacterial community undergoes dramatic changes during the first year of life, with different sets of taxa colonizing the skin at different timepoints. We confirmed our hypothesis that early life skin microbiome can be associated with later onsets of AD. We also confirm that the delivery mode influences the skin microbiome in the first day of life and that biological differences exist between different types of VD and CS. Our results suggest that while this impact appears transient, it may still play a role in the priming and development of the

ACK N OWLED G M ENTS
The authors thank all the study participants and all the individuals involved in facilitating and running the study. In particular, Julie Pernot, Louis Mercier, and Audrey Chuat for their contributions to the microbiome samples processing and sequencing. The PreventADALL study has been funded by the following public funding bodies: The Regional

CO N FLI C T O F I NTER E S T S TATEM ENT
EMR reports personal fees for presentations from Sanofi Genzyme, Novartis, Leo Pharma. MEDA and Omega Pharma outside the submitted work.