The association between the composition of the early-life intestinal microbiome and eczema in the first year of life

Introduction The early-life intestinal microbiome plays a crucial role in the development and regulation of the immune system. Perturbations in its composition during this critical period have been linked to the development of allergic diseases. Objective This study aimed to investigate the association between the composition of the early-life intestinal microbiome and the presence of eczema in the first year of life using shotgun metagenomic sequencing and functional analyses (metabolic pathways). Methods Stool samples from 393 healthy term infants collected at 1 week of age were analyzed with shotgun metagenomic sequencing. Environmental and clinical data were prospectively collected using 3-monthly validated questionnaires. Participants were clinically assessed during study visits at 12 months of age. Eczema was diagnosed by the UK diagnostic tool and by a research nurse. Data analysis was stratified by delivery mode. Results Eczema was diagnosed in 16.4% (60/366) of participants by nurse diagnosis. Infants born by cesarean section (CS) with nurse-diagnosed eczema had a higher relative abundance of Escherichia, Shigella, Enterobacter, and Citrobacter and a lower relative abundance of Veillonella than CS-born infants without eczema. In addition, CS-born infants without eczema had a higher abundance of genes involved in lactic fermentation. Vaginally born infants with eczema had a higher relative abundance of Bacteroides and a lower abundance of Streptococcus. Conclusion There is an association between the bacterial composition of the intestinal microbiome at 1 week of age and the presence of eczema in the first 12 months of life. Graphical Abstract


Introduction:
The early-life intestinal microbiome plays a crucial role in the development and regulation of the immune system. Perturbations in its composition during this critical period have been linked to the development of allergic diseases.
Objective: This study aimed to investigate the association between the composition of the early-life intestinal microbiome and the presence of eczema in the first year of life using shotgun metagenomic sequencing and functional analyses (metabolic pathways).
Methods: Stool samples from 393 healthy term infants collected at 1 week of age were analyzed with shotgun metagenomic sequencing. Environmental and clinical data were prospectively collected using 3-monthly validated questionnaires. Participants were clinically assessed during study visits at 12 months of age. Eczema was diagnosed by the UK diagnostic tool and by a research nurse. Data analysis was stratified by delivery mode.
Results: Eczema was diagnosed in 16.4% (60/366) of participants by nurse diagnosis. Infants born by cesarean section (CS) with nurse-diagnosed eczema had a higher relative abundance of Escherichia, Shigella, Enterobacter, and Citrobacter and a lower relative abundance of Veillonella than CS-born infants without eczema. In addition, CS-born infants without eczema had a higher abundance of genes involved in lactic fermentation. Vaginally born infants with eczema had a higher relative abundance of Bacteroides and a lower abundance of Streptococcus.

Introduction
Atopic eczema, also called atopic dermatitis, is an inflammatory disease of the skin characterized by a variety of symptoms, including skin dryness, itchiness, reddish patches, and xerosis. Eczema is considered the first manifestation of the atopic march, and infants affected by eczema are more likely to suffer from other allergic diseases later in life (van der Hulst et al., 2007;Hill and Spergel, 2018;Sroka-Tomaszewska and Trzeciak, 2021). Furthermore, severe eczema is associated with the presence of other diseases, such as diabetes, cardiovascular and autoimmune diseases, and mental health disorders (Sroka-Tomaszewska and Trzeciak, 2021).
Infants are the age group most frequently affected by eczema (de Lusignan et al., 2021). Risk factors for eczema include being first born, being of Asian or black ethnicity, having a higher socioeconomic status, and living in an urban area (Penders et al., 2013;de Lusignan et al., 2021). The intestinal microbiome also plays a crucial role in the development and regulation of immune responses. Consistent with this, there is evidence that allergic diseases, such as eczema, are influenced by the composition of the intestinal microbiome . As a critical step in developing immunotolerance, the immune system learns to recognize commensal microbiota. It has been hypothesized that reduced exposure to microbial immune stimulation, particularly via reduced microbiome diversity, impairs the healthy development of the immune system and increases the risk of allergic disease (Rook and Stanford, 1998). However, a recent systematic review suggests that particular microbes might be more important than bacterial diversity . Children who develop eczema, allergic sensitization, or asthma have been reported to have a greater abundance of Clostridiaceae and Enterobacteriaceae and a lower abundance of Lactobacillaceae . The findings reported from studies in neonates are more consistent than those reported from studies in older children and suggest that the early-life microbiome is crucial. This is the first study to use shotgun metagenomic sequencing and functional analyses to investigate whether the composition of the early-life intestinal microbiome is associated with the prevalence and severity of eczema in the first year of life in a large cohort of infants. Understanding this relationship might help with the early detection of infants at risk of developing eczema and provide a basis for evidencebased interventions to prevent or correct dysbiosis of the microbiome.

Study design and participants
The participants were a subset of infants from the Melbourne Infant Study: BCG for Allergy and Infection Reduction (MIS BAIR) . In this randomized trial, 1,272 healthy infants were recruited antenatally to investigate whether Bacille Calmette-Gué rin (BCG) immunization given in the first 10 days of life protects against childhood infection, allergy, and asthma. Inclusion criteria for infants were as follows: born after at least 32 weeks of gestation; birth weight greater than 1,500 g; absence of symptoms or signs of illness, including skin diseases; and no contraindication to BCG vaccination. Detailed data on demographics, pregnancy, and the perinatal period were collected at baseline. Three-monthly validated parent questionnaires were used to prospectively collect data on eczema-related symptoms, severity, and management. A visit by a study nurse took place when children were 12 months of age to clinically assess the incidence and severity of eczema.

Eczema diagnostic tools
The presence of eczema in the first 12 months of life was evaluated with two different complementary methods: (1) the cumulative incidence of any eczema in the first 12 months of life assessed by the UK diagnostic tool using data from the 3-monthly questionnaires (UK diagnostic tool) (Williams et al., 1994a;Williams et al., 1994b); and (2) the point prevalence of research nurse-diagnosed eczema at 12 months of age assessed by clinical assessment using the SCORing Atopic Dermatitis (SCORAD) system (nurse diagnosis).
The UK diagnostic tool was adapted in accordance with the participants' ages, as well as for online data collection (Pittet et al., 2022a). Eczema was defined as an itchy skin condition (major criterion) accompanied by at least three of the following minor criteria: (1) generally dry skin; (2) lesions in skin creases or cheeks; (3) atopic disease in a first-degree relative; and (4) visible eczema involving the flexures, head, or limbs. Previous data show that parents are able to accurately report eczema in their infants (Fleming et al., 2001).
In the absence of an agreed tool to reliably diagnose eczema in infancy (Pittet et al., 2022b), the use of two complementary tools improves the reliability of the results. Although the point prevalence estimate by a trained professional (nurse diagnosis outcome) is the most robust assessment, it captures lesions present on the day of visit only, depends on the management of eczema (i.e., effective treatment will mean lesions are absent), may miss seasonal fluctuation of eczema, and could capture lesions that are not necessarily atopic dermatitis (Pittet et al., 2022b). Being widely used, the UK diagnostic tool enables comparison with other studies; however, it is less relevant for infant eczema, with one of the major limitations being that the score requires the subject to experience itching (which excludes infants who are not yet developmentally capable of scratching themselves). In addition, the minor criteria are primarily based on flexural lesions, and not on the typical distribution of eczema in infancy (i.e., trunk and face, with only the latter included in the modified version) (Pittet et al., 2022b).
Age at onset of eczema, use of topical steroids, and eczema severity were also evaluated. Eczema severity was assessed using the Patient-Oriented Eczema Measure (POEM) score (Charman et al., 2004) included in each 3-monthly questionnaire and the SCORAD score no author, (no author, 1993) assessed by the research nurse at the study visit at 12 months.

Stool collection
Parents were asked to collect stool samples from their infants in sterile tubes on each of the first 10 days of life. To minimize variation, parents were asked to collect from the first bowel movement of the day (with date and time recorded) and immediately freeze the stool in their domestic freezers (-18°C or below). Samples were collected from parents' homes by a study team member and kept frozen during transportation to the laboratory, where they were aliquoted for storage at -80°C. One stool sample collected between days 5 and 7 of life was analyzed for each infant.

DNA extraction and shotgun metagenomic sequencing
Samples were thawed at room temperature and 100 mg of stool was mixed with 1x phosphate-buffered saline (PBS). DNA was extracted using the FastDNA ™ SPIN Kit for Soil (MP Biomedicals, Illkirch-Graffenstaden, France). Of this stool suspension, 300 µl was mixed with 978 µL of PBS (provided by the kit) and 122 µL of MT buffer and transferred into a lysing matrix E (LME) tube. Sample homogenization was performed with the FastPrep96 homogenizer using the program for human stool samples (settings: 1,600 rpm, 40 s). The LME tubes were centrifuged at 14,000 g for 10 minutes and the supernatant was mixed with 250 µL of protein precipitation solution in a 2-mL microcentrifuge tube. The tubes were inverted by hand 10 times and centrifuged at 14,000 g for a further 5 minutes. Subsequently, 1 mL of binding matrix solution was then added to the supernatant. The tubes were inverted for 2 minutes, then rested for 5 minutes, and 800 µL of the supernatant was removed. The remaining supernatant was resuspended in the binding matrix. At this point, 700 µL was transferred into a new spin filter tube and centrifuged for 1 minute at 14,000 g. This step was repeated twice. After this, 500 µL of ethanol-based washing solution (SEWS-M wash buffer) was mixed with the binding matrix. Samples were centrifuged three times for 1 minute at 14,000 g and allowed to dry at room temperature for 5 minutes. DNA was eluted in 50 µL of DNase-and pyrogen-free water.
The DNA concentrations of samples and a negative control (i.e., no sample DNA) were measured using Qubit ™ dsDNA High Sensitivity Assay kits (Life Technologies, CA, USA). The mean DNA concentration in samples was 367 ng; the negative control was below the detection limit (i.e., < 0.01 ng/µL) and was therefore not sequenced. Sequencing libraries were prepared with the Nextera DNA Flex library preparation kits (Illumina, San Francisco, CA, USA) to obtain insert sizes of approximately 600 bp. Illumina PhiX DNA was added to the libraries, which were then sequenced (2x149) on a NextSeq 550  Pre-processing, taxonomic assignment, and relative abundance computation The overall quality of the sequencing run was inspected with the Illumina Sequencing Analysis Viewer v. 2.4.7. The quality of the raw reads was assessed by FastQC v. 0.11.7 (http://www.bioinformatics. babraham.ac.uk/projects/fastqc/) and scanned with Trimmomatic v. 0.39 by using a sliding window of 20 nt (Bolger et al., 2014). Reads with an average quality below a Phred score of 28 and with a length of less than 100 nt (before or after trimming) were filtered out. Quality-filtered reads were then analyzed with Kraken v. 2.0.9-beta (Wood et al., 2019) and sequences mapping to the human (vGRCh38.p11) genome with a confidence score of more than 0.1 were discarded. The remaining reads were assigned to bacterial, archaeal, viral, and fungal genomes with Kraken2 (with a confidence score of more than 0.1) and with MetaPhlAn v. 3.0.7-1 (Segata et al., 2012). Compared with Kraken2, MetaPhlAn 3 uses a smaller reference sequences database, made up of clade-specific gene markers. Data obtained by MetaPhlAn 3 were used to confirm the results from the Kraken2/Bracken pipeline.
Read counts obtained with Kraken2 were corrected with Bracken v. 2.6.2 (Wood et al., 2019;Lu and Salzberg, 2020) at phylum, family, genus, and species levels. All the reference genome sequences were collected and analyzed with kmer2read_distr [settings: k-mer length (-k) of 35 and read length (-l) of 100] to finally generate a k-mer distribution file with generate_kmer_distribution.py. Eventually, the read abundance of each phylum, family, genus, and species in each sample was estimated with est_abundance.py (settings: minimum number of reads necessary for taxonomic classification (-t) = 10).
For each taxonomic level, the relative abundance was computed as a percentage by dividing the number of Bracken-corrected read counts in a given taxon by the total number of Bracken-corrected reads assigned to that taxonomic level in any given sample. MetaPhlAn 3 provided taxa and relative abundance already expressed as percentages.

Phages
The phage content was investigated by selecting reads assigned to viruses from the Caudovirales order. Caudovirales includes double-stranded DNA phages from the families Siphoviridae, Myoviridae, and Podoviridae, which are the most abundant bacteriophages present in the human intestine (Hoyles et al., 2014;Shkoporov et al., 2019). The relative abundance of phages was computed by dividing the number of reads mapped to phage genomes by the total number of quality-filtered no-human reads. The ratio was then multiplied by 1,000,000 and the relative abundance was expressed as counts per million (CPM).

Metabolic pathway analyses
Microbial metabolic pathways were identified by analyzing quality-filtered non-human reads with the HUMAnN v. 3.5 tool with default parameters (Beghini et al., 2021). Forward and reverse read files from each sample were concatenated in a unique FASTQ file. UniRef90 was used as a database. Normalized gene copies per million were then obtained from pathabundance output tables with human_renorm_table HUMAnN 3 code. The interpretation of metabolic pathways and their putative presence in each species were investigated by querying the MetaCyc database (https:// biocyc.org/) (Caspi et al., 2006).

Statistical analysis
All statistical analyses were performed with the R software v. 4.2.0. The codes used for statistical analyses are reported in Supplementary File 1.
To compare the bacterial compositions of the stool microbiomes at the species and genus level, principal coordinate analyses (PCoAs) were performed. For the PCoAs, Bray-Curtis dissimilarity distance was computed on the genera or species' relative abundance after square-root transformation with the vegdist function of the vegan R package v. 2.6-2. The resulting data were analyzed with the betadisper function (vegan). Differences in bacterial composition according to a given variable were assessed at the species, genus, family, and phylum taxonomic levels by permutational multivariate analysis of variance (PERMANOVA) tests. PERMANOVA was performed with the adonis2 function (vegan) using Bray-Curtis dissimilarity distance computed on square root-transformed data and using 9,999 permutations.
Ecological indices, richness, and Shannon diversity were computed after rarefying read counts analyzed with the Kraken2/Bracken pipeline to 35,000 with the rrarefy function (vegan). Shannon indexes were computed with the diversity function (vegan).
Prior to running differentially abundant analyses, we filtered out genera and species identified with the Kraken2/Bracken pipeline that had less than 10 reads in less than 100 (25.5%) of 393 samples. LEfSe was also used to identify differentially abundant pathways with the same settings as described above, except the lda_cutoff was set at 1.5. For all other statistical testing, the Wilcoxon rank-sum test was used. Results were considered significant when associated p-values (or false-discovery-rate-adjusted p-values for DESeq2 analyses) were less than 0.05. Plotting was performed with the ggplot2 package v. 3.3.6.
As delivery mode was expected to have a strong effect on the intestinal microbiome (Zimmermann and Curtis, 2018), data analysis was stratified according to delivery mode. Further factors investigated for their effect on the composition of the intestinal microbiome included infant (sex, ethnicity, gestational age, birth weight, feeding method, and antibiotic exposure), maternal (antibiotic use during the last trimester of pregnancy or during labor, and smoking), environmental (number of siblings, childcare attendance, number of smokers in the household, and animal exposure during pregnancy or at 12 months of age), and familial [family history of atopic disease (defined as eczema, hay fever, or asthma) or allergies (food and/or secondary allergies, eczema, hay fever, and/or asthma)] factors.

Ethics
Informed consent was obtained from participants' parents or guardians. The study was approved by the Royal Children's Hospital Human Research Ethics Committee (HREC, authorization 53543). MIS BAIR is registered with the Australia and New Zealand Clinical Trials Registry (1051228) and the US National Institutes of Health (NCT01906853).
There was no association between eczema prevalence and clinical variables such as delivery mode sex and family history (Supplementary File 2, Table 1).
Delivery mode, gestational age, formula milk, antibiotics during labor, and number of siblings had a strong effect on the bacterial composition at all taxonomic levels (PERMANOVA test, Table 3; Supplementary File 2, Tables 2-4). Birth weight had an effect on the composition of bacterial families, genera, and species (Table 3;  Supplementary File 2, Tables 3, 4). Concerning animal exposure, we did not find consistent results among the taxonomic levels: exposure to livestock at 12 months was significantly associated with bacterial composition at the phylum level only (Supplementary File 2, Table 2); exposure to pets (excluding cats and dogs) and livestock during pregnancy was associated with family, genus, and species composition (Table 3; Supplementary File 2, Tables 3, 4). Childcare attendance was significantly associated with phylum composition, as investigated by MetaPhlAn 3 (Supplementary File 2, Table 2). No associations were found between microbiome composition and sex, ethnicity, antibiotic exposure during the first year of life, smoking during pregnancy, antibiotic use during the last trimester of pregnancy, presence of smokers in the household at 12 months of age, cat and dog exposure during pregnancy or at 12 months of age, familial factors (presence of atopic disease or allergies in any parent and/or sibling, or presence of atopic disease in both parents), or use of topical steroids during the first 12 months of life.
No archaeal taxa were identified by MetaPhlAn 3. The only archaeal taxon detected by Kraken2\Bracken was the family of Methanobacteriaceae from the phylum Euryarchaeota. This family was detected in only two samples (read counts 12 and 28, respectively).
In total, 29 fungal species were identified by Kraken2\Bracken, belonging to 3 phylum, 21 families, and 21 genera. These species were identified in 68 (17.3%) samples and had read counts that did not allow further analyses (on average 0.03% of total non-human quality filtered reads). One fungal species (Candida albicans) was detected by MetaPhlAn 3 in three samples with a relative abundance between 0.02% and 0.29%. Because of the low abundance of archaea and fungi, the association between their composition in the intestinal microbiota and eczema outcomes could not be analyzed.

Intestinal bacterial microbiome composition and the presence of eczema
As expected, in PCoA plots, the stool microbiome profiles of CS-born infants clustered further apart from those of vaginally born infants than the stool microbiome profiles of infants with eczema  ) clustered from those of infants without eczema (Figure 1; Supplementary File 2, Figures 1-3). There was no difference in bacterial species diversity or richness between CS-born or vaginally born infants with or without eczema (Supplementary File 2, Figure 4). However, in CS-born infants with nurse-diagnosed eczema, the microbiome composition was significantly different at all taxonomic levels to that of CS-born infants without eczema (PERMANOVA test, p-values 0.0297, 0.0162, 0.0176, and 0.0125 for phylum, family, genus, and species, respectively; Supplementary File 2, Table 5).
Overall, the four most abundant phyla found in the intestinal microbiome were Firmicutes, Proteobacteria, Bacteroidetes, and Actinobacteria. No significant differences in the relative abundance of these phyla were detected in relation to eczema outcomes in either CS-or vaginally born infants (Supplementary File 2, Figure 5).
The 10 most abundant families were Enterobacteriaceae, Bacteroidaceae, Bifidobacteriaceae, Veillonellaceae, Streptococcaceae, Staphylococcaceae, Pasteurellaceae, Tannerellaceae, Clostridiaceae, and Enterococcaceae. Compared with CS-born infants without eczema, CSborn infants with eczema had a higher relative abundance of Enterobacteriaceae (p-values of 0.06 and 0.15 for the UK diagnostic tool and nurse diagnosis, respectively) and a lower relative abundance of Veillonellaceae (Bacillota) (p-values of 0.48 and 0.006 for the UK diagnostic tool and nurse diagnosis, respectively) (Supplementary File 2, Figure 6). Vaginally born infants with eczema had a lower relative abundance of Streptococcaceae than vaginally born infants without eczema (p-values 0.01 and 0.36 for the UK diagnostic tool and nurse diagnosis, respectively) (Supplementary File 2, Figure 6).
The 10 most abundant genera were Bifidobacterium, Veillonella, Bacteroides, Escherichia, Streptococcus, Klebsiella, Shigella, Phocaeicola, Staphylococcus, and Parabacteroides, and the 10 most abundant species were Escherichia coli, Bifidobacterium longum, Veillonella parvula, Parabacteroides di stasonis, Staphylococcus epidermidis, Haemophilus parainfluenzae, Streptococcus salivarius, Phocaeicola dorei, Bifidobacterium breve, and Shigella sonnei. Genera and species, which were identified by Kraken2/Bracken and had >10 reads in at least one-quarter of the sample cohort, were kept for further analyses. After this filtering step, 12.6% (111/882) of genera and 11.6% (326/2806) of species were retained. CS-born infants with eczema had a higher relative abundance of the genera Escherichia and Shigella than CS-born infants without eczema (p-values 0.049 and 0.008, respectively, for nurse diagnosis) and a lower relative abundance of Veillonella (p-value 0.004 for nurse diagnosis) (Supplementary File 2, Figure 7). Although not among the 10 most abundant genera, Enterobacter and Citrobacter were also more abundant in CS-born infants with eczema than CS-born infants without eczema. Vaginally born infants with eczema had a higher abundance of Bacteroides and a lower abundance of Streptococcus (p-values 0.045 and 0.011, respectively, for the UK diagnostic tool) (Figure 2; Supplementary File 2, Figures 7, 8).
The seven species Bacteroides xylanisolvens, Citrobacter youngae, Shigella boydii, Escherichia fergusonii, Shigella flexneri, Shigella dysenteriae, and Shigella sonnei were found to be discriminative for CS-born infants with eczema diagnosed by the UK diagnostic tool or nurse diagnosis (Figure 2), with a significantly higher abundance in infants with eczema. Seven Veillonella species, V. parvula, V. dispar, V. infantium, V. rogosae, V. tobetsuensis, V. denticariosia, and V. caviae, were found to be discriminative by LEfSe analyses, with a lower abundance in CS-born infants with nurse-diagnosed eczema (Figure 2; Supplementary File 3). Compared with vaginally born infants without eczema, Streptococcus rubneri and Streptococcus ilei were less abundant in vaginally born infants with both UKdiagnostic-tool-and nurse-diagnosed eczema.

Metabolic pathways differentially abundant in the intestinal microbiome and the presence of eczema
Eighteen pathways were significantly overrepresented in CSborn infants with both UK-diagnostic-tool-and nurse-diagnosed eczema. These were linked to: IQR, interquartile range. * eczema, hay fever, and/or asthma. ** food-and/or secondary allergies, eczema, hay fever, and/or asthma.  (1) amine and polyamine biosynthesis (superpathway of polyamine biosynthesis I); (2) amino acid synthesis (L-ornithine biosynthesis II and superpathway of arginine and polyamine biosynthesis, Lmethionine biosynthesis, L-homoserine and L-methionine biosynthesis, and S-adenosyl-L-methionine biosynthesis); (3) antibiotic resistance (polymyxin resistance) (4) carbohydrate degradation (L-rhamnose degradation I, chitin deacetylation, starch degradation III, and sucrose degradation II) (5) cofactor, carrier, and vitamin biosynthesis (superpathway of pyridoxal 5-phosphate biosynthesis and salvage, and pyridoxal 5phosphate biosynthesis I) (6) fatty acid degradation (fatty acid beta-oxidation I and superpathway of glyoxylate cycle and fatty acid degradation); and (7) Tricarboxylic acid (TCA cycle I, TCA cycle V, and superpathway of glyoxylate bypass). Pathways from amine and polyamine biosynthesis; amino acid synthesis; cofactor, carrier, and vitamin biosynthesis; and fatty acid degradation are present in Escherichia coli species according to the MetaCyc database. Genes involved in homolactic fermentation and biotin biosynthesis II were significantly overrepresented in CS-born infants without eczema when compared with CS-born infants with eczema, irrespective of diagnostic tool (Supplementary File 2, Figure 9). Functional analyses of the vaginally born cohort did not result in overlapping findings between the UK diagnostic tool and nurse diagnosis.

Intestinal phage composition and the presence of eczema
In total, viruses were detected in 379 infants (96.4%) and accounted for 0.2% (IQR 0.05%-0.48%) of quality-filtered nonhuman reads. To investigate phages, species belonging to the Caudovirales order were selected. These were the only detected viruses in 359 samples (91.3%). The most abundant detected phages were those infecting Bacteroides, Enterococcus, Escherichia, Enterobacteria, Streptococcus, and Staphylococcus. No significant differences were detected in the composition of phages according to eczema outcomes (Supplementary File 2, Figure 10).

Discussion
In this study, in the first week of life, CS-born infants who subsequently developed eczema had a significantly different microbiome composition at all taxonomic levels from CS-born infants without eczema. Compared with the UK diagnostic tool, nurse diagnosis at 1 year of age is a more robust assessment as it captures infants with more severe and persistent eczema. This explains the larger differences in the composition of the intestinal microbiome observed between infants with and without nursediagnosed eczema compared with the differences observed using the UK diagnostic tool, as the latter also captures infants who might have only transitorily suffered from mild eczema.
Consistent with our findings, previous studies have also reported that infants who develop eczema have a higher relative abundance of Enterobacteriaceae, especially Escherichia, Shigella, and Klebsiella, in their intestinal microbiome (Penders et al., 2006;Penders et al., 2007;Hong et al., 2010;Ta et al., 2020). Furthermore, a higher relative abundance of Enterobacteriaceae, especially Escherichia, Shigella, Enterobacter, and Citrobacter, have been found in the intestinal microbiome of children who develop food allergies and asthma (Gosalbes et al., 2013;Azad et al., 2015;Zheng et al., 2016;Tanaka et al., 2017). Similar to our findings, a lower relative abundance of Veillonella in the intestinal microbiome has been found to be associated with the development of atopic disease and food allergies in children (Arrieta Species microbiome profiles according to delivery mode and eczema outcomes. Principal coordinate analyses (PCoAs) were performed on the relative abundance of 2,806 species detected in 393 infants with the Kraken2/Bracken pipeline. Axes indicate the percentage of total variance. Colors indicate delivery mode (A) and delivery mode and eczema outcomes (B, C). For eczema outcomes, participants with missing values were excluded from the analysis and plotting; thus, (B) was computed on 91.6% (360/393) of samples and 98.3% (2,757/2,806) of species, and (C) on 93.1% (366/393) of samples and 98.5% (2,765/2,806) of species. The centroid indicates average microbiome species composition. Lines connect centroids to corresponding data points. Chen et al., 2016;Tanaka et al., 2017). In contrast, one small study investigating the intestinal microbiome and the prevalence of eczema simultaneously at 1 year of age found a positive association between the relative abundance of Veillonella and eczema (Zheng et al., 2016). This could be explained by the intestinal microbiome being analyzed at 1 year of age rather than 1 week of age.
Bacteroides, especially Bacteroides fragilis, have previously been reported to be more abundant in the intestine of infants who develop eczema (Vael et al., 2008;Vael et al., 2011;Wang et al., 2016). We found this association only in infants who were vaginally born. This was likely to be because of the low abundance of Bacteroides in the CS-born infants, which is a common observation in infants born by CS (Zimmermann and Curtis, 2018). Immunoglobulin (Ig) A is produced in mucosal tissue, including at high concentrations in the intestine. IgA promotes intestinal barrier function and is involved in the maintenance of host-commensal mutualism (Fagarasan et al., 2010). The binding of IgA to Bacteroides thetaiotaomicron inhibits the innate immune response (Peterson et al., 2007). Bacteroides are the main producer of the short-chain fatty acid propionate (Louis and Flint, 2017) and have been associated with both increased and decreased inflammation of the intestine (Hertli and Zimmermann, 2022). Propionate increases glucose-6phosphatase activity in the jejunum, which induces gluconeogenesis (Kimura et al., 2011;De Vadder et al., 2014;Chenard et al., 2020). Increased glucose levels in stool samples have previously been found in infants who develop eczema (Ta et al., 2020). However, in the study that found increased glucose levels, infants who developed eczema were reported to have delayed colonization by B. fragilis. It has been hypothesized that the increased glucose levels result from a depletion of bacterial groups with the ability to derive butyrate and propionate through direct or indirect pathways, which results in reduced glycolysis (Ta et al., 2020). Differentially abundant bacterial species. Bar plots show the log10-transformed linear discriminant analysis (LDA) score for a given species obtained by LEfSe analysis according to the delivery mode and the presence of eczema diagnosed with a given tool. LDAs report which are the major species explaining the differences in microbiome species relative abundance between infants with and without eczema. Positive values of LDA correspond to species that are overrepresented in infants with eczema; negative values to those without eczema. * = species that were also detected as differentially abundant with DESeq2 (FDR-adjusted p-value < 0.05). Bold = species that were found to be differentially abundant by both eczema diagnosis tools in CS-or vaginally born infants. Values are also reported in Supplementary File 3.
A lower relative abundance of Streptococcus in the intestinal microbiome has previously been associated with the development of eczema in children (Zheng et al., 2016;Yu et al., 2021). In our study, we found a lower relative abundance of Streptococcus in both CSand vaginally born infants with eczema; however, the difference was statistically significant only in the latter. This could be because a higher proportion of infants born by CS received formula milk, which has been associated with a lower abundance of Streptococcus (Zimmermann and Curtis, 2018).
Previous studies have reported an association between a lower relative abundance of Bifidobacterium and the development of eczema (Watanabe et al., 2003;Murray et al., 2005;Mah et al., 2006;Hong et al., 2010;Penders et al., 2010). Although Bifidobacterium was among the most abundant genera identified in our study, we did not observe such an association. Similarly, Clostridioides difficile has previously been found to be more abundant in the first month of life in infants who develop eczema (Penders et al., 2007;van Nimwegen et al., 2011), an association we also did not find in our study. In mice, Clostridia-related bacteria called segmented filamentous bacteria are reported to promote interleukin (IL)-6 and IL-26 production by dendritic cells, which leads to the induction of T helper (Th) 17 cell differentiation (Ivanov et al., 2009). Th17 cells produce pro-inflammatory cytokines such as IL-17A, IL-17F, and IL-22, which have been associated with the development of autoimmune disease (Littman and Rudensky, 2010). Clostridium spp. cluster IV and XIVa strains, as well as B. fragilis, have been associated with an increase in IL-10producing T regulatory (Treg) cells (Atarashi et al., 2011;Round et al., 2011). Treg cells accumulate in the intestine and play a critical role in the maintenance of immune homeostasis and the development of autoimmunity and allergies (Turner et al., 2020).
A higher relative abundance of Lactobacillus paracasei in the early intestinal microbiome has been found to reduce the risk of developing eczema (Penders et al., 2010). Administered as a probiotic, Lactobacillus rhamnosus, which belongs to the same taxonomic group as L. paracasei, halves the risk of developing eczema (Wickens et al., 2008). In in vitro studies, exopolysaccharides from L. rhamnosus have been shown to stimulate TNF, IL-6, and IL-12 secretion from human peripheral blood mononuclear cells (Chabota et al., 2001).
In our study, a lower abundance of genes involved in lactic fermentation was found in CS-born infants with eczema. Veillonella metabolize lactate, the final product of lactic fermentation, into the short-chain fatty acids acetate and propionate (Ng and Hamilton, 1973). These metabolites have anti-inflammatory activities and can help to maintain the function of the intestinal barrier (Nurmi et al., 2005;Tedelind et al., 2007;Suzuki et al., 2008). It remains unclear whether lactate has pro-or anti-inflammatory properties (Manosalva et al., 2022). It is also not yet known whether a lower abundance of Veillonella is associated with higher levels of lactate and whether this is associated with an increase in intestinal inflammation. Other bacteria that can ferment lactate, such as Lactobacillus, Streptococcus, and Bifidobacterium, have also been found to be less abundant in infants who develop eczema (Watanabe et al., 2003;Murray et al., 2005;Mah et al., 2006;Hong et al., 2010;Penders et al., 2010;Zheng et al., 2016;Yu et al., 2021).
Although it has been reported that being born by CS and receiving formula milk increases the risk of developing eczema (Kull et al., 2005;Yu et al., 2015), the evidence for this is not robust. In fact, recent studies and meta-analyses report that infants born by CS do not have an increased risk for developing eczema (Renz-Polster et al., 2005;Bager et al., 2008;Polos and Fletcher, 2019). Similarly, a protective effect of breastfeeding against developing eczema is not certain (Elbert et al., 2017;Wang et al., 2017;Bigman, 2020). In line with this, there is only a partial overlap between the differences found in the intestinal microbiome after being born by CS or receiving formula milk and the differences in the composition of the intestinal microbiome associated with the development of eczema. Infants born by CS are known to have a higher relative abundance of Clostridioides and Enterobacteriaceae other than Escherichia (e.g., Citrobacter and Enterobacter) in their intestinal microbiome, while the relative abundance of Escherichia has been reported to be lower (Zimmermann and Curtis, 2018). Furthermore, infants born by CS have a lower relative abundance of Bacteroides and Bifidobacterium, while the findings on the abundance of Streptococcus remain contradictory (Zimmermann and Curtis, 2018). A higher relative abundance of Veillonella has been found in infants born by CS (Hesla et al., 2014). Similarly, breastfed infants have been reported to have a higher relative abundance of Bifidobacterium and a lower relative abundance of Bacteroides and Clostridioides, while the findings for Streptococcus and Veillonella are contradictory (Zimmermann and Curtis, 2018).
Although preterm-born infants are known to have a higher relative abundance of Enterobacteriaceae (especially Escherichia, Enterobacter, and Citrobacter) and Bacteroides, as well as a lower relative abundance of Bifidobacterium, in their intestinal microbiome (Zimmermann and Curtis, 2018), they have not been found to develop eczema more frequently than term-born infants (Gerner et al., 2022). However, in addition to the findings above, preterm-born infants have also been found to have a higher relative abundance of Streptococcus and Veillonella (Zimmermann and Curtis, 2018), which might protect them from eczema. Infants exposed to antibiotics have been reported to have a significantly higher risk of developing eczema and other allergic diseases (Duong et al., 2022). Although infants exposed to antibiotics have been reported to have a lower relative abundance of Bifidobacterium and a higher relative abundance of Enterobacteriaceae in their intestinal microbiome, they have also been found to have a lower relative abundance of Bacteroides, while findings for Clostridioides remain contradictory (Zimmermann and Curtis, 2018). However, the changes in the intestinal microbiome induced by antibiotics are influenced by many other factors, such as the condition leading to the antibiotic use, underlying disease, hospital admission, and, even more importantly, the dose, duration, administration route, and spectrum of the antibiotic . Exposure to pets has been associated with a higher relative abundance of Veillonella (Azad et al., 2013), which might partially explain why children exposed to pets are less prone to eczema (Epstein et al., 2011). However, again, the association between pet exposure and a decreased risk of eczema is not clear (Langan et al., 2007).
In conclusion, our study of a large cohort of infants, using shotgun metagenomic sequencing and independent bioinformatic and statistical pipelines, found that the bacterial composition of the intestinal microbiome at 1 week of age is associated with the subsequent presence of eczema in the first year of life. Our study also identified putative genera (Escherichia, Shigella, Enterobacter, Citrobacter, and Bacteroides) that might play a role in the development of eczema and others (Streptococcus and Veillonella) that seem to be protective.
The limitations of our study include the fact that the bioinformatics analyses identified taxa by mapping reads to known reference genomes without using de novo assembly. Furthermore, reads were mapped individually in each sample. In future, new approaches based on assembling metagenomes by pooling reads from all samples might improve the analyses of the composition of the microbiome, while also enabling the recovery of the genome of community-shared strains that are not yet discovered (Uritskiy et al., 2018;Churcheward et al., 2022;Krakau et al., 2022). The DNA extraction protocol used was not specifically designed to target viral particles, which means that the abundance of bacteriophages might have been underestimated. Also, many intestinal phages are yet to be characterized (Sutton and Hill, 2019).
Differences between the findings in our study and other studies investigating the association between the intestinal microbiome and the risk of developing eczema might be explained by (1) differences in the investigated time points, i.e., 1 week of age might be too early to detect significant differences in the abundance of Lactobacillus, Bifidobacterium, and Clostridioides, which are more prevalent later in infancy; (2) differences in the definition of eczema outcomes; (3) differences in the methodologies used to investigate the microbiome; and (4) smaller cohort sizes.
Future studies should include multiple sampling time points to investigate the trajectory of the intestinal microbiome that is associated with the development of eczema and atopic disease in general. Furthermore, the exact mechanistic roles of individual bacterial species and the role of the interaction between these species in atopic diseases should be addressed. Findings from these studies will form the basis for evidence-based interventions to modulate the intestinal microbiome in infants and children at risk of eczema and atopic disease, such as modifying breast milk and intestinal microbiota with directed prebiotics and probiotics, and the use of bacteriophages.

Data availability statement
FASTQ files containing forward and reverse quality-filtered non-human reads were deposited in the European Nucleotide Archive (https://www.ebi.ac.uk/ena) under the project PRJEB55714. Codes used for the preprocessing of sequencing reads (quality-filtering, removal of human DNA, etc.) are deposited at GitHub: https://github.com/miraclelabunifr/Eczema.

Ethics statement
The studies involving human participants were reviewed and approved by the Royal Children's Hospital Human Research Ethics Committee HREC, authorisation (53543). Written informed consent to participate in this study was provided by the participants' legal guardians/next of kin.

Author contributions
PZ conceptualized and supervised the study. NC is the chief principal investigator of the MIS BAIR trial. WJ performed the sequencing. SL, PZ, LP and NM helped with data cleaning. SL, OC, LP, PZ, and LF analyzed the data. SL and PZ wrote the initial draft of the manuscript. SL, OC, LP, LF, NC, and PZ revised the manuscript. All authors contributed to the article and approved the submitted version.