Atopic dermatitis (AD) is one of the most common inflammatory skin diseases and clinically manifests as an eczema-like rash with pruritus (Ständer, 2021; Zhang et al., 2023). The prevalence of AD is 15–20% among children and up to 10% among adults, making AD the 15th most common nonfatal disease and the skin disease with the highest disease burden in disability-adjusted life years (Ständer, 2021; Schuler et al., 2023). In addition, AD may be associated with an increasing prevalence of psychosocial disorders, such as depression, anxiety, sleep disorders and suicidal ideation (Kage et al., 2020). Epidermal barrier dysfunction, skin microecological imbalance and immune disorders, which are dominated by type 2 inflammation, contribute to the outbreak of AD (Langan et al., 2020).

The skin microbiota has a complex interaction with the host (Chen et al., 2018). In recent years, with the application of high-throughput DNA sequencing technologies based on two principal methods (16S rRNA gene sequencing and metagenomics sequencing), the study of the AD skin microbiome has shown great progress (Byrd et al., 2018; Ghosh et al., 2018). Skin microbiota disturbances characterized by increased Staphylococcus aureus colonization and decreased microbiota diversity occurred in both lesion and nonlesion sites of AD patients (Koh et al., 2022). S. aureus may destroy the skin barrier and induce an inflammatory response (Langan et al., 2020). The local Th2 immune response further diminishes barrier function and facilitates the growth of Staphylococcus, especially S. aureus (Langan et al., 2020; Schuler et al., 2023). However, the composition of the skin microbiome is complex, and the specific mechanism of its interaction with the host remains unclear.

Metabolomics is a technique that reflects the phenotypic outcome of biological activities and elucidates the relationship between metabolite changes and physiological/pathological changes (Zhang et al., 2023). Skin cells and glandular secretions, skin resident microbiota, and external environmental factors (such as cosmetics or pollution) can affect the composition of skin metabolites (Afghani et al., 2022). The metabolomic profile of 15 AD patients and 17 controls was studied through skin punch biopsies, which revealed that a total of 77 metabolites differed significantly between the lesional skin of atopic dermatitis, nonlesional skin of atopic dermatitis and skin of controls (Ilves et al., 2021). Until now, much of the metabolomics research on AD skin has focused on the proportion of changes in the lipid composition of the skin due to its role in the structural integrity of the skin barrier (Emmert et al., 2021; Zhang et al., 2023).

Theoretically, the skin microbiome and the skin would constantly interact with surrounding skin metabolites. The identification of AD skin metabolites may help to determine the changes in host status and the influence of the skin microbiome on the local skin. Several previous studies have explored the interactions between the skin microbiota and skin metabolites in AD. AD skin has an increase in free fatty acids (FFA), and this effect appears to be S. aureus-dependent (Emmert et al., 2021). Saturated shorter-chain FFA are negatively correlated with Staphylococci (Emmert et al., 2021), and long-chain FFA are decreased in S. aureus-colonized AD skin (Li et al., 2017). Shorter fatty acids can more easily traverse the skin to acidify it, and S. aureus does not grow well in acidic healthy skin pH conditions (Afghani et al., 2022). However, lipids are not the only metabolites present in the skin, and an increasing number of studies are starting to turn from lipids to other molecules in AD (Afghani et al., 2022). Herein, we performed 16S rRNA gene sequencing of the skin microbiome and untargeted metabolomics analysis of AD patients’ skin samples, and searched for metabolites closely related to the skin microbiome of AD patients through association analysis. We also explored the effects of sex and rash status on the results through subgroup analysis. We established a correlation map of the skin microbiome and skin metabolome in AD patients and healthy people. These results highlight the role of the skin microbiome in regulating overall metabolism and provide new insights into the pathological mechanisms of AD.

Skin microecological imbalance plays an important role in the pathogenesis of AD (Langan et al., 2020). The relationship between human and skin microbiota is bidirectional. Skin microbiota can affect host gene expression, and different skin characteristics can also form microbiota with specific preferences (Fyhrquist et al., 2019). There are also complex interactions between different skin bacteria. Previous studies have suggested that S. aureus is often dominant in the skin of AD patients, which can promote the onset of AD and limit the growth of other bacteria, which may have regulatory or protective effects (such as Staphylococcus epidermidis and Corynebacterium) (Fyhrquist et al., 2019). In contrast, some strains of Staphylococcus epidermidis, Roseomonas and Propionibacterium can inhibit the growth of S. aureus and are potential probiotics for the treatment of AD (Koh et al., 2022). Our study is based on 16S high-throughput sequencing technology, and the results are difficult to accurately annotate to the species level, nor can they be used to detect viruses or eukaryotic communities (Ghosh et al., 2018). Our study showed that the alpha diversity of the skin microbiota decreased in AD patients. The relative abundance of the genus Staphylococcus significantly increased, while the relative abundances of the genera Propionibacterium, Brevundimonas, Lactobacillus and Lysobacter significantly decreased, suggesting that the genus Staphylococcus affected the growth of other skin symbiotic flora and destroyed the skin microecological balance. Thus, it is reasonable to infer that S. aureus should account for most of the detected genus Staphylococcus and play an important role in the pathogenesis of AD.

In addition to disease factors, the skin bacterial structure is also affected by other confounding factors. It has been reported that the number of bacteria at the genus level does not change from birth to 1 year of age, but the relative abundance of microbiota fluctuates, with the relative abundances of the genera Staphylococcus and Streptococcus decreasing as infants grow, while the relative abundances of other genera increase (Capone et al., 2011). This trend was maintained through childhood (2–12 years old), with more diversity of skin microbiota on the forearms of children than adults (Shi et al., 2016). The composition of skin microbiota in adolescents was stable, similar to that of adults (Shi et al., 2016). The skin microbiome structure in different sites of AD was also different (Bjerre et al., 2021). Therefore, the subjects included in this study were all adults, and the sampling sites were all elbows, which can better reduce the interference of age and site factors on the research results. To investigate the effect of sex and local rash factors on the results, we performed a subgroup analysis. We revealed that there were significant differences in the composition and structure of the bacterial community between different sexes and lesion states.

Due to the important role of skin microbiota in the pathogenesis of AD, restoring skin microbial homeostasis is becoming a novel therapeutic strategy for treating AD (Hrestak et al., 2022; Koh et al., 2022). Application of prebiotics and probiotics, whether by gastrointestinal approach or external use, may help to increase the diversity of the skin microbiota and have potential therapeutic effect of AD (Hrestak et al., 2022). Coagulase-negative Staphylococcus (CoNS) strains with antimicrobial activity were common on the normal population but rare on AD subjects, and application of antimicrobial CoNS strains to human subjects with AD may decreased colonization by S. aureus (Nakatsuji et al., 2017). Topical microbiome transplantation with Roseomonas mucosa for AD patients was associated with significant decreases in measures of disease severity and S. aureus burden (Myles et al., 2018). A Living symbiotic bacteria-involved skin dressing was developed for microbiome-based biotherapy toward AD, which may recover skin barrier functions and alleviate AD-associated inflammation responses (Liu et al., 2023).

Samples for metabolomics analyses can be serum, urine, sweat, skin, etc. (Afghani et al., 2022). Both host gene expression products and substances produced by skin microorganisms influence skin metabolites (Afghani et al., 2022). Skin samples can be collected by invasive or noninvasive methods; the former is by skin biopsy, and the latter is by tape strips (Afghani et al., 2022). It has been reported that the swabbing method is comparable to the tape stripping method for collecting viable skin bacteria without losing fidelity to the composition of the skin microbiome (Ogai et al., 2018). However, to our knowledge, there have been no previous studies in AD metabolomics applying the swabbing method for sampling. In this study, a polyester swab was used to collect samples from skin, and large amounts of metabolites were successfully obtained. This noninvasive method for skin sample collection was proven to be safe, convenient and efficient.

Diet, age, sex, sampling method and other factors may affect the metabolites detected (Afghani et al., 2022). We evaluated the influence of disease, sex and rashes on the metabolomics analyses. The PCA results revealed that metabolites of healthy people have better consistency than those of AD patients, and males have better consistency than females. The number of differentially abundant metabolites per capita suggested that the effects of the state of AD, sex and the presence or absence of rashes on metabolomics analyses were successively reduced.

The differentially abundant metabolites between AD patients and HVs were enriched in the purine metabolism and phenylalanine metabolism pathways. The contents of xanthosine/xanthine and L-phenylalanine/trans-cinnamate are significantly elevated in AD, especially among females. However, we should notice that the metabolites we detected do not necessarily derived from microbes, they may also origin from the host. Xanthine is the substrate of xanthine oxidase and xanthine dehydrogenase, and it is also the intermediate product of the process from hypoxanthine to uric acid (Kulikowska et al., 2004). Xanthine is involved in a variety of intracellular metabolic pathways, such as purine nucleotide catabolism (Kulikowska et al., 2004). Exogenous guanosine and xanthosine, which are fluxed through the GTP branch of purine biosynthesis, were shown to significantly reduce methicillin-resistant S. aureus (MRSA) β-lactam resistance (Nolan et al., 2023). Further study demonstrated that exposure of MRSA to guanosine and xanthosine can significantly reduce the levels of the cyclic dinucleotide c-di-AMP, which is needed for β-lactam resistance (Nolan et al., 2023).

Phenylalanine is involved in a variety of metabolic activities in the body, and tyrosine, phenylactate and trans-cinnamate can be produced under the action of different enzymes (Matthews, 2007; Teufel et al., 2010; Otto et al., 2019). Bacteria in the environment can participate in the metabolism of phenylalanine and phenylacetate (Teufel et al., 2010). Pseudomonas taiwanensis has been used to catalyze trans-cinnamate formation from phenylalanine (Otto et al., 2019). The functional predictions based on metagenomic analysis in previous study revealed that AD skin samples exhibited enrichment in phenylalanine tyrosine and tryptophan biosynthesis (Chng et al., 2016). In addition, human lack the ability to synthesize essential amino acids such as phenylalanine, and they acquire essential amino acids from their diet or perhaps their associated microbial communities (Shrode et al., 2022; McCann and Rawls, 2023). Essential amino acid may act as chemical mediators of host–microbe interaction (McCann and Rawls, 2023). Thus, it was reasonable to infer that the significant proportion of increasing phenylalanine in AD we detected by skin metabolomics was probably derived from dominant microbes in AD. As far as we know, there are no relevant studies on purine metabolism and phenylalanine metabolism pathways in the pathogenesis of AD, but there is evidence that xanthosine, phenylalanine and trans-cinnamate participate in bacterial metabolic activities (Otto et al., 2019; Nolan et al., 2023). The important role of S. aureus in the pathogenesis of AD has been generally confirmed, and this study suggested that xanthosine/xanthine and L-phenylalanine/trans-cinnamate were positively correlated with Staphylococcus. Therefore, the purine metabolism and phenylalanine metabolism pathways may play a certain role in the pathogenic mechanism of S. aureus in AD, which is worthy of further exploration.

Epidermal barrier dysfunction is also a major characteristic of AD (Langan et al., 2020). Ceramide plays an important role in maintaining skin barrier function, and sphingosine is an important component of ceramide (Toncic et al., 2020). Ceramides can be divided into several subtypes, and they are dysregulated in AD patients (Ghosh et al., 2018). Previous studies have observed a decrease in long-chain ceramides and an increase in short-chain ceramides in AD patients (Afghani et al., 2022). It was reported that the colonization of Staphylococcus was positively correlated with the epidermal ceramide subspecies AS, ADS, NS and NDS in AD (Emmert et al., 2021). Our study showed that the amount of sphingosine in AD patients was approximately 10 times that in healthy controls (p < 0.05), which was consistent with previous studies (Toncic et al., 2020). However, we did not identify ceramide substances in this study, which may be due to the wide variety of ceramides and lack of a specialized ceramide comparison database.

There are some limitations in our study. First, the correlation between the skin microbiota and skin metabolites is only statistically significant, and it was difficult to determine whether the differential metabolites between the two groups detected in skin samples came from the host or microorganisms. Second, our study only detected the metabolites in skin samples without untargeted metabolomics analysis on serum samples, which can better reflect the metabolic situation of the body. It is necessary to collect both skin samples and blood samples in future research. In addition, relatively fewer subjects were included in the subgroup analysis, which inevitably increased the random error. Thus, the specific mechanism of interaction between skin microbiota, human skin and skin metabolites still needs to be further studied.

In summary, AD patients had dysbiosis of the skin microbiome with the features of decreased species richness and evenness. The relative abundance of the genus Staphylococcus increased significantly in AD, while the relative abundances of the genera Propionibacterium, Brevundimonas, Lactobacillus and Lysobacter were significantly decreased. Multiple metabolites related to purine metabolism and phenylalanine metabolism pathways (such as xanthosine/xanthine and L-phenylalanine/trans-cinnamate) were upregulated in AD patients, and they were positively correlated with the genus Staphylococcus, which suggested that purine metabolism and phenylalanine metabolism pathways may play a certain role in the pathogenic mechanism of S. aureus in AD. We also found that different sexes had certain effects on skin microbiota and metabolite composition. Females are more likely to be colonized by the genus Staphylococcus than males, and the differentially abundant metabolites involved in purine metabolism and phenylalanine metabolism pathways were more obvious in female.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found in the article/Supplementary material.

The studies involving humans were approved by the Ethics Committee of Guangdong Provincial Hospital of Chinese Medicine (BE2019-165-01). The studies were conducted in accordance with the local legislation and institutional requirements. The participants provided their written informed consent to participate in this study.

KH: Software, Writing – original draft, Visualization, Funding acquisition, Formal analysis. FL: Writing – original draft, Investigation, Data curation. YLiu: Writing – original draft, Investigation, Data curation. BL: Data curation, Writing – original draft, Investigation. PQ: Writing – original draft, Data curation. LY: Writing – original draft, Formal analysis. SH: Writing – original draft, Formal analysis. WL: Writing – review & editing, Project administration, Funding acquisition, Formal analysis. XM: Writing – original draft, Funding acquisition, Data curation. LD: Writing – review & editing, Supervision, Project administration, Funding acquisition, Conceptualization. YLin: Writing – review & editing, Supervision, Resources, Project administration, Funding acquisition, Conceptualization.