**4. Discussion**

We examined the diversity and community structure of both skin and gut (feces) microbiota in young infants with allergies symptoms, using 16S rRNA gene amplicon sequencing. In addition to being the first of its type to be performed on Polish patients, the study has four key strengths. Firstly, the gut and skin microbiota were examined simultaneously and compared. Secondly, a novel analysis was performed on the skin microbiota in infants with FA. Thirdly, the study group was characterized by a range of clinical pictures, which allowed the differences in the microbiotas to be highlighted. Finally, unlike most previous studies, the participants were under six months of age; this allows the relationship between the microbiota and the development of allergy to be assessed in an early stage of life.

The bacterial communities found in human gut and skin vary considerably between individuals and with regard to developmental stages, as do views on their relationship with allergy development [13,15,16,22,26]. These differences might stem from the profile of the studied group, e.g., the number of patients, clinical picture, demographics and environmental conditions, as well as on the methodology used, such as the primer system, PCR conditions, sequencing technology and the database used for classification [51]. This variability might explain why our results disagree with many published studies.

However, similarly to other studies [3,23,26,32,52–54], no significant differences in bacterial alpha diversity were found between allergic and non-allergic infants, regardless of the type of allergy or compartment, i.e., skin or feces. These results suggest that, contrary to results obtained for older patients [29,55], decreased diversity does not appear to be related to allergy symptoms, even in the case of AD. This might be due to the fact that the microbial community is still developing in young infants; nevertheless, this observation indicates that neither species richness nor diversity can be used as predictors of allergy in infants under one year of age. However, in contrast to our present findings, a number of previous studies have reported significant variation in gut microbiota α-diversity between infants with FA and healthy children [25,56]. It is possible that this variation could be attributed to the differences between the groups of participants: age (10 months [25] vs four months in our study), feeding method (currently breastfeeding 39% [25] vs 59% in our study), type of allergy (IgE-mediated [25] vs IgE + nonIgE-mediated (56 and our study)), type of allergenic food (egg [25] vs milk in our study) or, finally, genetic and environmental difference (China [56] vs Europe in our study).

We observed a significant correlation between fecal bacterial community structure and allergy type, even though the communities varied greatly within one group. This result is similar to those obtained previously [3,12]. Interestingly, both skin and fecal microbial communities differed significantly between the AD and ADFA groups, suggesting that these two clinical conditions differ in their immunological background; this difference is known to influence the microbiome, and vice versa [13,15]. Skin communities also differed, suggesting that assessing skin swabs could be a diagnostic tool for particular phenotypes of allergy. It could be possible to devise a classifier (e.g. using RandomForest framework) [57] to predict the type of allergy based on community structure. Such a tool could be useful in clinical practice, aiding in diagnosis. However, this will require further studies based on larger numbers of patients, i.e., providing greater statistical power.

Some of the taxa whose abundance was found to differ between allergic and healthy infants have previously been found to be related to allergies, although the direction of these relations varied. We found that the abundance of Bacteroidales and their member taxa, down to the level of genus (*Bacteroides* and *Parabacteroides*) in the libraries derived from feces, was related to certain types of allergy. This relationship might be driven by the production of the epithelial barrier-breaching agent phenol/*p*-cresol, which is known to be produced by the members of the Bacteroidaceae [58]. Previous studies have noted an increase in Bacteroidales members correlated with various types of allergy [53,59,60]. On the other hand, contrary to our present findings, a number of studies have reported a higher abundance of Bacteroidetes in healthy individuals [11,20,56]; however, these studies employed groups comprised of participants from different countries (Canada, Taiwan, China) and at different ages (thirteen months [20] vs three months, twelve months [11], three months [56], four months in our study).

Bacteria of the order Xanthomodales, particularly those of the genus *Xanthomonas*, are frequently found on healthy skin, albeit at a low abundance [61]. In the present study, they were associated with skin communities from allergic patients: the first such report. The abundance of *Xanthomonas* was found to increase after successful treatment of AD with emolients [62]. Although the precise role of these bacteria in skin dysbiosis remains unclear, it is possible that their keratinolytic properties may be involved. Alternatively, they might participate in shaping the microbiome structure via microbe–microbe interactions.

Many authors have indicated that Staphylococcus plays an important role in the relationship between AD and skin colonization [29,32,63]. Kong et al. reported an increase in the number of Staphylococcus, including both *S. aureus* and *S. epidermidis*, in patients aged 2–15 years with AD exacerbation, and that topical skin treatments increased the proportion of Streptococcus, Propionibacterium and Corynebacterium [29]. Similarly, our findings indicated that the children with AD were characterized by a smaller proportion of Streptococcus than the controls.

Assuming the existence of the gut–skin axis in humans, i.e., changes in the skin microbiome that have been found to modulate the gut microbiome [64], and changes in skin physiology that appear to induce changes in the gut microbiota [65], we hypothesize that microbiota of the skin would influence that of the gut and vice-versa. This relationship was examined in the present study.

It seems likely that the gut microbiome influences the health status of skin [66–69], particularly in the pathogenesis of AD [17]. Certain organisms, such as *Escherichia*/*Shigella* or *Veillonella*, were found to be more abundant in the feces of AD patients, while others (*Streptococcus*, *Haemophilus*) were more prominent in those of healthy subjects [69].

Our observations suggest that gut dysbiosis might be involved in the development of atopic skin inflammation, even though different organisms were differentially abundant in feces and skin libraries. Similar observations have been reported previously [23,33,35,54,66,69].

In spite of the great variability in bacterial communities observed in our samples, we found certain bacterial taxa to be characteristic of particular clinical conditions, and identified a positive correlation between the abundance of certain taxa (*Gemella*, *Lactobacillus*, *Bacteroides*) in the skin samples with those in the fecal samples. This observation, made for the first time, supports the skin–gut axis hypothesis [16–18,64,66].

Certain taxa found to be correlated between compartments in the present study have previously been noted as having an influence on allergy, e.g., *Bacteroidetes* [53] and lower taxa, particularly the genera *Bacteroides* and *Parabacteroides* in the intestines [60,70] or the genus *Staphylococcus*, particularly *S. aureus*, in the skin [29,32,63]. The taxa that correlated between compartments differed depending on the clinical picture, which could be explained by two factors: (i) transmission of microorganisms from the skin to the guts and (ii) selection by that immune system. Indeed, Beijerinck notes, "everything is everywhere, but the environment selects", or rather, microbial communities are shaped by a combination of environmental filtering and stochasticity [71,72]. Therefore, our observed correlation indicates that selection operated similarly in the compartments in question.

Hence, we hypothesize that the key factor affecting the community assembly of the skin and feces is the immunological profile of the host. If the organisms co-vary, then the immune system exerts similar pressure on them in different compartments, and their abundance depends on the initial load. As the sets of strains differ between healthy and allergic children, it is possible that the two groups demonstrate different immune responses. Such correlations may well be of practical importance and serve as indicators of the patient's state of health. Interestingly, although the correlated organisms did not predominate, they were nevertheless found to be involved in allergy pathogenesis [73–75].

Our findings indicate that the abundance of the genus *Gemella* was positively correlated between the feces and skin communities in all allergic children, but not in healthy ones. No similar observation was found in the literature, but some relationships between *Gemella* and allergic inflammation seems to be possible. Taylor et al. [73] report a positive correlation between *Gemella* and sputum eosinophil percentage in patients with asthma.

One of the OTUs whose abundance was correlated between the gut and the skin was *Bifidobacterium scardovii*; however, this was only observed in the allergic group. This could be explained by the fact that although most of the bacteria belonging to the genus *Bifidobacterium* have been recognized as beneficial to human health, *Bifidibacterium scardovii* is the one of five strains of the *Bifidobacterium* which are not—its presence has been associated with clinical conditions such as urinary infections [76].

Among the other OTUs with a positive correlation between fecal and skin microbiota in the allergic patients was *Corynebacterium nuruki*. Generally, *Corynebacteria* are believed to have a rather positive impact on human health [77–79]; however, *Corynebacterium kroppenstedtiian* has been associated with the microbiome of lesioned skin in AD patients [75] and the abundance of the genus *Corynebacterium* has been associated with chronic rhinosinusitis [74]. A novel finding in the present study was that OTU20, identified as *Rothia mucillaginosa*, also demonstrated a positive correlation between the skin and feces of allergic patients. Surprisingly, a positive correlation was also demonstrated for OTU 38 and 92, belonging to the genus *Lactobacillus*, which is known to bestow a protective effect against allergy [80].

Despite its strengths, our study has also some limitations. Most importantly, the group was quite small. This is particularly the case for the AD group, which can be attributed to the selection bias associated with recruiting the patients in an academic center specialized in FA management. In addition, despite the wealth of experience of the participating center, FA remains difficult to diagnose in such young infants; therefore, continuation of the study is planned to confirm our findings by diagnosing the children when they are older. The study group was also heterogeneous with respect to certain important characteristics, such as breastfeeding, introduction of solid food or probiotic intake; however, no statistically significant differences were found between the study and control groups with regard to these variables. It was also heterogeneous with respect to the severity of symptoms at the time of sample collection. Due to the non-specificity and diversity of allergy symptoms in such young infants, it would be very difficult to collect a group with identical clinical parameters.

Another weakness of the study lies in the selection of primers; they could be not well-suited for both compartments, and some organisms, such as *Staphylococcus*, might not be well classified. Skin microbiota is usually assessed based on the V1-3 region and it is not encouraged to sequence V4 only [81]. On the other hand, our system generated a long fragment (450 bp), allowing for classification down to the species level for most organisms. It was also found that the V3-V4 region yielded better results in terms of diversity recovered than the V1-V2 region [82]. Finally, we collected swabs from healthy skin because not all children were in the acute phase of AD at the time of enrollment; however, we think that this approach was justified as Leung et al. report significant differences between the microbiota of non-lesioned, apparently healthy skin of children with AD and ADFA, but no such difference was observed for tissue displaying a skin rash [63].

Understanding the interrelationship between the skin and the gut may allow the design of new treatment models for skin diseases based on the manipulation of the gut microbiota, and conversely, for gastrointestinal diseases through the manipulation of skin microbiota. Moreover, further research on the microbiome may allow the determination of specific patterns of bacteria in multiple body compartments, which are prognostic factors for the development of allergies, and various clinical symptoms (AD, FA). They can also lead to the selection of specific probiotics for treating allergic conditions, either applied to the skin or applied enterally; our analysis suggests that *Acinetobacter* spp., *Haemophilus haemolyticus*, *Bacteroides ovatus*, *Schaalia odontolytica* and *Actinomyces graevenitzi* may be potential candidate bacteria for such treatment, their numbers being correlated with the feces and skin of healthy children.
