Gut microbiota profile in children affected by atopic dermatitis and evaluation of intestinal persistence of a probiotic mixture

Atopic dermatitis (AD) has been hypothesised to be associated with gut microbiota (GM) composition. We performed a comparative study of the GM profile of 19 AD children and 18 healthy individuals aimed at identifying bacterial biomarkers associated with the disease. The effect of probiotic intake (Bifidobacterium breve plus Lactobacillus salivarius) on the modulation of GM and the probiotic persistence in the GM were also evaluated. Faecal samples were analysed by real-time PCR and 16S rRNA targeted metagenomics. Although the probiotics, chosen for this study, did not shape the entire GM profile, we observed the ability of these species to pass through the gastrointestinal tract and to persist (only B. breve) in the GM. Moreover, the GM of patients compared to CTRLs showed a dysbiotic status characterised by an increase of Faecalibacterium, Oscillospira, Bacteroides, Parabacteroides and Sutterella and a reduction of short-chain fatty acid (SCFA)-producing bacteria (i.e., Bifidobacterium, Blautia, Coprococcus, Eubacterium and Propionibacterium). Taken togheter these results show an alteration in AD microbiota composition with the depletion or absence of some species, opening the way to future probiotic intervention studies.

However, recommendations on timing and dose of administration of probiotics in AD have not yet been established. Moreover, not all the studies based on probiotic intake demonstrate the actual ability of probiotics to colonize the GT. This aspect does not allow clinicians to choose the probiotic strain with the certainty that it persists in the GM. In the present study, GM of AD patients and healthy age-matched controls were profiled to identify bacterial biomarkers associated with the disease. The effect of the intake of probiotics composed by Bifidobacterium breve and Lactobacillus salivarius was evaluated with respect to GM modulation over time to assess the persistence of the probiotic bacteria by quantitative Real-Time PCR (qRT-PCR).

Materials and Methods
Study design and sample collection. Nineteen patients in the age range of 0 to 6 years with a diagnosis of AD were prospectively enrolled in the study at the Dermatology Unit of the Bambino Gesù Children's Hospital in Rome, Italy. Patients took the probiotic (composed of B. breve BR03 and L. salivarius LS01) twice per day (1 × 10 9 UFC/dose of each species for 20 consecutive days). Exclusion criteria consisted of treatment with steroids or calcineurin inhibitors, antibiotics intake or gastrointestinal disorders in the four weeks before enrolment and during the follow up. Faecal samples were collected during clinical visits at time T 0 (no probiotic intake), T 1 (completion of probiotic intake), T 2 (60 days after the end of probiotic intake); T 3 (90 days after the end of probiotic intake).
The treatment was based on antiseptics, emollients, hydration and also focused on therapeutic patient education. In only two patients it was necessary to recurr to topyc steroids during the last two time points.
Seventy-two faecal samples were collected and accompanied by a clinical data diary, SCORAD index values, comorbidities, vaccinations and diet type. The samples were stored at −80 °C at the Human Microbiome Unit of Bambino Gesù Children Hospital in Rome until DNA extraction. Eighteen faecal samples from healthy children of the same age range were also collected as age-matched controls (CTRLs). The criteria for the CTRLs were the absence of chronic diseases or gastrointestinal infections and no antibiotic or probiotic intake in the four weeks before the enrolment. This study was approved by the OPBG Ethics Committee (protocol number 391LB).
All research was performed in accordance with relevant guidelines/regulations. Informed consent was obtained from all parents and/or legal guardians of participants.

Isolation of B. breve BR03 and L. salivarius LS01 bacterial strains from the probiotics. The
B. breve BR03 and L. salivarius LS01 bacterial strains were cultured on Columbia agar +5% sheep blood medium (COS, Biomerieux Marcy l'Etoile, France) at 37 °C for 24 h under anaerobic (B. breve) or aerobic (L. salivarius) conditions. Single colonies were isolated and purified on new COS plates based on their morphology. Bacterial identification was confirmed using a matrix assisted laser desorption/ionization mass spectrometry (MALDI-TOF MS) biotyper and a Microflex LT mass spectrometer (Bruker Daltonics, Bremen, Germany) as described 15 .
Bacterial DNA cloning and species-specific primer and probe design. The purified PCR products were sequenced and cloned into the PGEM vector following the instructions provided by the pGEM ® -T Easy Vector System kit (Promega, Italy) using Escherichia coli competent cells as a host. The obtained plasmids, pGEM-BB (pGEM + B. breve) and pGEM-LS (pGEM + L. salivarius), were extracted (Plasmid Miniprep Kit, Promega, Italy), quantified using the NanoDrop ND-1000 spectrophotometer and diluted. The dilutions, which ranged from 10 6 to 10 1 vector copy numbers, were used as standards in the quantitative RT-PCR (qRT-PCR) assays. The cloned fragments of 16S rDNA in the pGEM-BB and pGEM-LS vectors were amplified and sequenced with an automated sequence analyser (Genetic Analyser 3500, Applied Biosystems, CA, USA) using a 50-cm capillary array and a POP-7 polymer (Applied Biosystems) and the BigDye Terminator Cycle Sequencing kit (Applied Biosystems, version 3.1) according to the manufacturer's instructions. All electropherograms were manually edited for base ambiguity. The obtained FASTA sequences were aligned using CLUSTAL-W software (http://www.ebi.ac.uk/clustalw/) and used for the design of species-specific primers and TaqMan probes (Roche Diagnostics, Mannheim, Germany) (Supplementary Table 1).
Bacterial DNA extraction from stool samples. Frozen stool samples were thawed at room temperature, and DNA was manually extracted using the QIAmp Fast DNA Stool mini kit (Qiagen, Germany) according to the manufacturer's instructions. DNA was quantified using the NanoDrop ND-1000. Comparable amounts of DNA (80 ng) from each sample were used in the qRT-PCR assays.
Targeted-metagenomics. The V1-V3 regions (520 bp) of the 16S ribosomal RNA (rRNA) locus were amplified and pyrosequenced using a 454-Junior Genome Sequencer (Roche 454 Life Sciences, Branford, USA) 17 . The obtained raw reads were analysed using Quantitative Insights into Microbial Ecology software (QIIME) 1.8.0 software 18 by demultiplexing, quality score checking, low-length excluding and denoising 19 . Sequences were grouped into operational taxonomic units (OTUs) by clustering at a threshold of 97% pairwise identity using UCLUST for sequence clustering 20 and the representative sequences were submitted to PyNAST for sequence alignment 18 . The Greengenes database (v 13.8) was used for OTU matching.  Table 1), 3 µl molecular-grade H 2 O and 5 µl DNA template. Each sample was tested in duplicate to ensure data reproducibility. The RT-PCR temperature profile consisted of an initial denaturation at 95 °C for 10 min, 45 amplification cycles at 95 °C for 10 sec, 60 °C for 30 sec and 72 °C for 1 sec followed by a final cooling step at 40 °C for 30 sec. Absolute quantification was performed using the "second derivative maximum method" 21 .
Statistical analyses. All data were tested for normal distribution using the Shapiro-Wilk normality test.
Statistical analyses were computed using the phyloseq R package for alpha and beta diversity 22 . Furthermore, the adonis function in the vegan R package was used to perform the PERMANOVA test on beta diversity with 999 permutations using the "strata" argument within the adonis function.
The non-parametric Mann-Whitney U-test and Wilcoxon signed-rank test were used to compare the two independent groups (CTRL versus AD at T 0 ) and the data for the time points within the AD group, respectively. The linear discriminant effect size (LEfSe) was computed 23 with α value equal to 0.05 and a logarithmic LDA score threshold of 2.0. The area under the relative operating characteristic (AUROC) test and discriminant analysis (DA) based on univariate ANOVA, Fisher's coefficient and leave-one-out classification were applied. Correlations between B. breve and L. salivarius concentrations were determined by the Spearman test using IBM SPSS statistical software (version 21). Only significant p-values (p < 0.05) corrected using the Holm method were considered 24 .

Results
This study included 19 AD patients and 18 CTRLs, age 0 to 6 years. Patient and CTRL characteristics are summarized in Table 1. In particular 12/19 were vaginal delivered, 13/19 were breast-fed and 15/19 were weaned after 6 months.
Targeted-metagenomics. Eighty-eight faecal samples were analysed by 16S targeted-metagenomics with two samples excluded due to the poor quality of the metagenomic reads, which consisted of 18 baseline (T 0 ) and 52 follow-up (19 at T 1 , 17 at T 2 , 16 at T 3 ) samples from the AD patients and 18 from the CTRL subjects. From the total set of samples, 239,153 sequencing reads with a mean value of 2,202 sequences for each sample were obtained.
The microbiota biodiversity of the AD and CTRL groups was determined using alpha and beta diversity analyses. In the comparison between the AD and CTRL groups, we tested the effects of the clinical variables (i.e., delivery, neonatal feeding before weaning, weaning time, age, and BMI) on the AD sample distribution. Beta diversity analysis (Bray-Curtis, unweighted and weighted UniFrac metrics) which included all these variables, showed that "age" and "weaning time" greatly influenced the distribution of samples in the principal coordinate analysis (PCoA). Patients under the age of one year (infants) differed enormously from the rest of the sample cohort and formed a significantly separated cluster ( Supplementary Fig. 1, Supplementary Table 2). To avoid bias in the comparison between the AD and CTRL groups, we excluded patients under the age of 1 year, thus eliminating the age and weaning time effects (Supplementary Table 3). Hence, final analysis included 15 AD (patients older than 1 year of age) and 18 CTRLs. www.nature.com/scientificreports www.nature.com/scientificreports/ To evaluate how OTUs were differentially distributed in the AD and CTRL groups, differences in beta diversity were calculated. A clear separation was observed between the two groups ( Fig. 1), which was verified using the PERMANOVA test (p = 0.001 for weighted, unweighted and Bray-Curtis analyses). However, there were no clearly defined clusters for the patients stratified into T 0 and T 1 -T 3 groups (PERMANOVA > 0.05) ( Supplementary Fig. 2).
Alpha diversity was calculated with respect to OTU richness, evenness and rarity to understand the ecological differences within the AD and CTRL groups using the Shannon, observed and Chao1 indices, respectively. AD patients showed a significantly lower level of alpha biodiversity according to the observed and ChaoI indices compared to the CTRLs at each time point (T 0 , T 1 , T 2 , T 3 ) (Fig. 2, Supplementary Table 3). In addition, the Shannon index revealed less biodiversity in the AD patients, but this result did not reach statistical significance. No statistically significant differences amongst T 0 -T 3 time point paired comparison were achieved (Supplementary Table 3).
To detect differences in OTU composition between AD patients and the CTRLs, we compared time point T 0 of the AD group (i.e., before the probiotic intake) to the CTRL group. The OTU distribution was investigated at the phyla, family and genus levels. At the phylum level, the Mann-Whitney U-test highlighted the OTU abundance differences, which showed that Bacteroidetes was significantly higher in the T 0 AD group and Actinobacteria and TM7 were significantly increased in the CTRLs (Fig. 3, Panel A; Supplementary Table 4).
At the family level, the AD group was characterised by lower relative abundances of the Actinobacteria families, such as Propionibacteriaceae, Actinomycetaceae, Coriobacteriaceae and Bifidobacteriaceae, and the complete absence of Corynebacteriaceae (Fig. 3, Panel B; Supplementary Table 5). Moreover, the Firmicutes families also showed different distributions between the two groups. In particular, Erysipelotrichaceae and Enterococcaceae were substantially reduced in AD patients, and Staphylococcaceae was completely absent. In contrast, these patients were highly enriched in Bacteroidaceae (Bacteroidetes) compared to the CTRLs, which accounted for up to 23% of the overall families followed by the Proteobacteria family Alcaligenaceae (2.6%) (Fig. 3, Fig. 3).
To understand the 16 OTUs belonging group, we applied the average area under the relative operating characteristic (AUROC) test. According to the classification proposed by Swets (Swets 1988), AUROC values ranging www.nature.com/scientificreports www.nature.com/scientificreports/ from 0.7 to 1 were considered accurate to discriminate between the groups. This analysis showed that the 16 OTUs had discriminatory power. In particular, Bacteroides, Parabacteroides, Faecalibacterium, Oscillospira and Sutterella were selectively associated with the AD T 0 group, and Actinomyces, Propionibacterium, Bifidobacterium, Eggerthella, Staphylococcus, Enterococcus, Blautia, Coprococcus, Bulleidia, Corynebacterium and Eubacterium were discriminatory for the CTRL group (Fig. 4, Supplementary Table 7).  www.nature.com/scientificreports www.nature.com/scientificreports/ To estimate the statistical power of these OTUs to act as a classifier for each group, we used the DA model. The DA revealed that 96.9% of the original groups and 84.4% of the cross-validated groups were correctly classified (Supplementary Table 8).
To evaluate the influence of probiotic intake on gut microbiota modulation, we tested the OTU distribution using the pairwise Wilcoxon signed-rank test to compare all time point samples. The test did not find any significant differences at any taxonomic level.
RT-PCR analysis. RT-PCR was performed on AD patient samples (15 T 0 , 15 T 1 , 14 T 2 and 11 T 3 from >1-year-old patients) using primers and probes specific for B. breve and L. salivarius. The median concentrations (molecules/ul) of the two species were higher at T 1 than the other time points (Fig. 5). B. breve persisted until the latest time point (T 3 ) whereas L. salivarius decreased to zero by time point T 2 (Fig. 5). In particular, the B. breve concentration differences were statistically significant for the pairwise comparisons between T 0 and all of the follow-up time points. In contrast, L. salivarius concentration achieved significant differences between T 1 and all the other time points (Supplementary Table 9).   www.nature.com/scientificreports www.nature.com/scientificreports/ Discussion GM dysbiosis has been shown to precede the onset of AD 25 . However, studies on atopic diseases and microbiota are conflicting because both decreased and increased bacterial community diversity have been related to atopy 26 . Our results were consistent with previous findings of low intestinal microbial diversity in AD [27][28][29] , which supports the theory of 'microbial deprivation syndromes of affluence 30 . According to this theory, reduced intensity and diversity of microbial stimulation lead to an abnormal immune maturation in early childhood. In fact, limited microbial pressure results in insufficient Th 1 cell induction and the failure to suppress Th 2 responses. The switching of the immune stimulation towards a pronounced Th 2 -phenotype is suggested to be a major mechanism to explain allergy development and maintenance 30,31 .
Faecalibacterium genus is generally an indicator of the healthy status of the gut due to its anti-inflammatory effects 36,37 . Low levels of Faecalibacterium and in particular of F. prausnitzii in the GM have been associated with Crohn's disease 38 and AD children 33 . A possible explanation of the high levels of this genus in AD, reported by Song and co-workers 37 , is that the inflamed epithelium with a barrier dysfunction, which is typical in AD, can release nutrients that stimulate the growth of F. prausnitzii subspecies that are not short-chain fatty acid (SCFA)-producing. The decreased production of SCFAs, such as butyrate, could lead to further inflammation in the gut epithelium.
Oscillospira is a common inhabitant of the GM. It is a butyrate-producing bacterium that utilizes host glycans as growth substrates and contributes to the maintenance of gut health 39 . The role of Oscillospira in atopy is not clear. However, our results show that it is strongly associated with AD, which is consistent with the results of Canani and co-workers 32 . Its presence in the GM of AD children could likely be linked to the high abundance of Bacteroidaceae, which produce fermented products that serve as substrates for Oscillospira growth 40 .
Bacteroides spp. are common inhabitants of the human gut, however their increased presence has been associated with food allergy and other atopic manifestations [41][42][43] . Indeed, higher levels of Bacteroides in atopy could lead to the continuous production of lipopolysaccharides (LPS), the major component of gram-negative cell wall, in the gut, which could trigger an inflammatory response 44 . Moreover, Bacteroides species are reported to alter gut permeability 41,45 , a condition observed in AD.
Among the Proteobacteria, Sutterella levels were increased in AD children. This genus has been associated with other inflammatory diseases, such as Crohn's disease and ulcerative colitis, but also with healthy adults 46,47 . Thus, it is still not completely clear if Sutterella is involved in inflammation or is a normal inhabitant of the human GM.
Our results revealed that AD GM is characterised by low colonisation of OTUs that have a role in the maintenance of gut health, like Actinomyces and Eggerthella.
Actinomyces spp are members of the normal oral microbiota and have been reported as one of the causative bacteria of dental caries and periodontal diseases 48 . Several studies reported that Actinomyces spp. are early colonisers of the healthy infant gastrointestinal tract (GT) 49 .
Eggerthella belongs to the Coriobacteria that are common members of the human GM 50 . They are assacharolytic bacteria that produce formate and lactate only from glucose 49 . Until now, only Eggerthella lenta and the still not fully characterized Eggerthella spp. YY7918 are associated with the human gastrointestinal tract 49 . Morinaga (1988) studied the role of E. lenta in the stimulation of the immune system and found that this bacterium is involved in the production of anti-tumour molecules that stimulate natural killer cells 51 . Furthermore, E. lenta could have a role in the stimulation of hepatic detoxification and in the inactivation of pharmaceuticals in the gut 52 .
Moreover, AD patients showed a strong reduction of some SCFA-producing bacteria, such as Bifidobacterium, Blautia, Coprococcus, Eubacterium and Propionibacterium 53,54 . SCFAs are of particular interest for maintaining host health because they may exert anti-inflammatory effects through several mechanisms, including epithelial integrity (preserving tight junctions) and maintenance of the mucus layer 55 . The resulting low production of SCFAs may be causative of the intestinal barrier dysfunction, increased intestinal permeability and inflammation found in ADs 56 . In particular, Coprococcus spp. are butyrate-producing bacteria 49 . Interestingly, Nylund and co-workers reported an inverse correlation between the SCORAD index and the levels of Coprococcus eutactus, confirming the role of this particular genus in the amelioration of AD 29 .
Bifidobacterium spp. are assumed to be beneficial for human health due to their several effects such as vitamin production, immune system stimulation, inhibition of potentially pathogen bacteria, improvement of food ingredients digestion 57,58 . In the contest of allergic diseases, several studies based on murin and in vitro models, have higlighted the potential role of Bifidobacterium in reducing inflammation by inducing the production of anti-inflammatory cytokines and suppressing Th 2 immune response and IgE production 25,59,60 . The absence of Bifidobacterium in AD children is consistent with other studies 35,61 and could lead to a lack of anti-inflammatory effects.
Existing treatments for AD are limited. Therefore, the focus is now to identify alternative strategies. Based on the hygiene hypothesis, probiotics have been proposed as therapeutic and preventive interventions for allergic diseases with the aim to attenuate inflammation 62 . The most used species, belonging to Bifidobacteria and Lactobacilli, have been shown to reduce the risk of AD 63 .
Because of controversial results 64,65 , the influence of probiotics on the prevention or management of AD requires further investigation. In particular, the recommendations on the time of administration and the dose of the probiotic in AD have not been yet drawn up. Huang and collegues 13 reported a metanalysis of the data available on the topic, highlighting a multitude of administration approaches, including different doses and intake timing. For this reason we investigated the persistence of the two probiotic strains in the GT at defined time points. In our study, we observed an increase of B. breve and L. salivarius in the stool samples starting from time point T 1 . The significant increase of these species at the end of the probiotic intake, indicated the survival of the bacteria www.nature.com/scientificreports www.nature.com/scientificreports/ during their transit in the GT. Interestingly, B. breve persisted until the last time point (T 3 ), while L. salivarius did not, suggesting a species-specific survival in the GT.
Furthermore, investigating the action of the probiotic on the AD microbiota profile, we did not observe a significant change in the composition of the GM in AD children after the probiotic intake.
In conclusion, although the probiotics chosen for this study, did not shape the entire GM profile, we can confirm the ability of these species to pass the GT and to persist (only B. breve) in the GM. Moreover, our study revealed that the GM of children with AD is characterised by a dysbiotic status with a prevalence of some species such as Faecalibacterium, Oscillospira, Bacteroides, Parabacteroides and Sutterella, that can act as possible biomarkers associated to the disease. We also identified a reduction or complete absence of some microbes (i.e., Bifidobacterium, Blautia, Coprococcus, Eubacterium and Propionibacterium) with anti-inflammatory effects or involved in immune homeostasis, which might have a protective role against AD. Differences in microbiota composition between AD and CTRL could suggest to take into consideration, in future intervention studies, the species depleted or absent in AD gut microbiota as potential probiotic candidate.