October 8, 2026 | A study published July in Allergy (DOI: 10.1111/all.70449) is one of the largest and most thorough longitudinal studies to date assessing how the infant skin microbiome changes in conjunction with atopic dermatitis (AD, more commonly referred to as eczema), food sensitization (FS), and food allergy (FA). Using a sophisticated total DNA metagenomic sequencing approach, researchers showed that specific microorganism levels are already changed on the skin of healthy 2-3-month-old infants who will later develop eczema, but only when that eczema is combined with FS or FA. The researchers found “specific microbes that can help us distinguish the infants who would later develop both AD and FA compared to those who would develop AD alone and also from those who would not develop any of these diseases at 12 months,” according to corresponding first author Zeyang Shen, Assistant Professor in the School of Molecular Biosciences at Washington State University. Diagnostics World News spoke with Shen to learn how microbial analysis might improve the diagnosis, prediction, and treatment of childhood allergic conditions.
With allergic disease, there is a somewhat predictable sequence of symptom progression starting in childhood, beginning with eczema that leads to FA, allergic rhinitis (hay fever), and finally, asthma. This process is known as the atopic march, and it often begins at a few months of age and progresses to asthma by ages 3-8. Up to 20% of children experience eczema, but not all cases progress through other conditions. Additionally, while useful clinical shorthand, variations exist, including FA in isolation as an example. Genetics explain less than 25% of eczema and FA susceptibility. Variants of the FLG gene encoding fillagrin, an epidermal barrier protein, are known to be strongly correlated with both AD and FA, but other factors apparently contribute. The skin microbiome shows correlated dysregulation, but prior work has looked primarily at individuals only after AD or other allergic conditions were detected. In one notable exception, a study from 2023 performed longitudinal 16S sequencing of bacteria on infant skin at 3, 6, and 12 months of age, finding that early commensal Staphylococcus colonization appeared protective against progression to AD.
Shen and colleagues wanted to analyze microbiome development more thoroughly. They turned to the Australian VITALITY study, initially conducted to assess the effects of vitamin D supplementation on FA and AD development in nearly 2,800 infants. The primary result of that study, published February as a late-breaking abstract from the American Academy of Allergy, Asthma & Immunology annual meeting, was that supplementation did not significantly reduce FA on the whole, although the results suggested protective effects against milk and egg allergy. The full peer-reviewed study, which will include the secondary endpoint of AD development, has not been published.
A collaboration formed between VITALITY directors and Julia Segre’s group at the National Human Genome Research Institute, where Shen conducted his postdoctoral research focused on early-life skin microbiome genomics. For this study, an allergic-disease-enriched subset of 429 VITALITY infants was included. Importantly, 76 of these produced samples at both 2-3 months age, before any sign of disease, and at 12 months, enabling longitudinal analysis. “This is probably one of the largest and most precious infant cohorts today in terms of capturing really early onset of these atopic diseases, including AD, FA, FS, and it’s very rare that we have these samples from even before these diseases occur, at two to three months,” Shen said.
“VITALITY had very detailed clinical phenotyping, including physician-assessed AD, SCORAD [a measure of AD severity], FS, and challenge-confirmed FA, along with longitudinal sampling,” Shen explained. The one-year-olds were assessed for AD and food-specific IgE responses to any of 10 common allergens—egg, peanut, cow’s milk, cashew, sesame, shellfish, almond, soybean, hazelnut, or wheat. FS indicated food tolerance, whereas FA included clinical reactivity.
Microbiome Analysis: Sequence Everything
Skin swab samples were acquired from both the cheek and the inner elbow of infants, recording whether 12-month-old elbow samples came from an active eczema flare at acquisition. Unlike prior microbial analyses that focused on 16S ribosomal, bacteria-specific sequencing, shotgun metagenomic sequencing was used, reading every piece of DNA present, including fungi, viruses, and host genome.
The researchers first looked at the 12-month-olds, subdividing them into healthy controls, AD only, AD + FS, and AD + FA. AD-inclusive elbow samples were further divided into active or inactive eczema. Considering these subsets of AD with coexisting conditions was critical in identifying distinct populations. “I think the main point of this study is if we only look at all AD cases, all the signals have been mixed together, and it’s not really distinguishing any of those,” Shen said.
At the highest level, none of these disease conditions drastically changed the microbial communities. Measuring beta diversity, how different one communal population is from another—e.g., an eczema-inflamed elbow sample vs. a healthy control—Shen said that “we didn’t really see clusters of these samples; we don't even see clear clusters in terms of disease groups.” In contrast, beta diversity changes significantly between samples based on infant age or skin site. Among disease groups and healthy controls, when comparing within one skin site and age, the “forest” looks about the same then. These conditions “are not producing a wholesale restructuring of the skin microbiome in the way that something like a severe C. difficile infection can dramatically reorganize the gut microbiome,” Shen said. But down in the underbrush, subtle changes were afoot.
For instance, “If we take a look into AD alone, we saw the staph epi (Staphylococcus epidermidis) pops up,” Shen said, but this only occurred for that subgroup and only if the site had active eczema. “If we look at AD co-occurring with FA or FS, these other bugs pop up.” In AD with FS, while the S. epidermidis levels were indistinguishable from those of healthy controls, four other bacterial species’ levels were significantly decreased, but again, only on AD-affected sites. When AD combined with FA, the most significant changes were seen, with increased levels of two bacteria, decreased levels of four bacteria—distinct from the four reduced in AD + FS—and increased levels of one fungus, Malasezzia slooffiae. Only in the case of AD with FA were some of the changes preserved in cheek samples and AD-unaffected elbow samples. There was very little overlap in significant changes from healthy controls among the groupings. Put another way, Shen said, “there’re different microbes associated with AD alone versus AD plus FA or FS.” Additionally, there were no significant species-level changes for FS or FA in the absence of AD.
From here, the researchers looked back at the 2-3-month-old samples. With knowledge of which healthy children would go on to develop AD and/or FS/FA, could the nascent microbiomes portend these conditions?
The answer is nuanced. For AD in isolation, no significant correlation showed up; 2-3-month-olds who would go on to acquire AD looked no different than those who would not. However, correlations began to appear for infants who also developed FA or FS. The effect was most striking for AD + FA, where 4-5 specific bacterial species were enriched, generally at both elbow and cheek sites. Interestingly, none of these species were common with those significantly changed at one year; the trajectory of microbial alteration remains unclear.
Because their method also captures significant regions of host genomic DNA, Shen and colleagues considered host FLG mutations. Although loss-of-function FLG variants have been strongly associated with AD and FA, how they influence the microbiome is less well studied. Comparing the microbiomes of healthy infants with normal or mutant FLG revealed no significant changes. However, looking only at the AD group experiencing an active eczema flare, they saw 18 different species decreased for FLG mutation carriers relative to FLG-normal infants. “Genetics can interact with microbiome in terms of what kind of microbes we saw on the skin of those patients even within AD,” Shen explained. For the AD group without active eczema, none of these changes occurred, and in line with having no active eczema, the altered microbiome as a function of FLG status did not yet appear in 2-3-month-old samples. “Rather than thinking of host genetics and the microbiome as independent risk factors, there may be interactions in which host genetics alters the skin environment, which in turn favors particular microbial species or strains,” Shen said.
Functional Microbial Genes
In further demonstration of the power of metagenomic analysis, the researchers looked all the way into specific genes within specific microbes. “One of the things I find particularly interesting about this study is that the ‘microbiome’ is not just about which microbial species are present; we can keep zooming in,” Shen explained. “Even two infants carrying the same microbial species may carry genetically different strains, and some of those strain-level genetic differences may potentially matter biologically.”
They demonstrated a few examples. In one, within active eczema samples, missense mutations in the prs gene (related to nucleotide biosynthesis) of S. thermophilus were associated with more severe eczema. That is even more interesting because total S. thermophilus levels were not significantly changed in any comparison. This strain-level microbial genetic analysis provides a new avenue for discovery in infant skin and allergic conditions. However, the result is correlational for now. “It is an association that gives us a biological hypothesis to investigate: perhaps strains with different metabolic capabilities interact differently with the host or behave differently in the inflamed skin environment,” Shen said.
Finally, maternal skin microbiomes were analyzed for a subset of infants. “We found that mothers and infants frequently shared strains of the same species—for example, in eight of nine families in which both mother and infant carried S. aureus, they shared the same strain,” Shen said. “That suggests maternal transmission can help establish part of the infant skin microbiome.” S. aureus was one of the strains correlated at 2-3 months with later AD + FA development, and although some of the strains persisted through this development, S. aureus levels were not correlated with any of the subgroups at 1 year of age. Other studies have associated S. aureus with more severe AD, but the AD samples in this study were predominantly mild.
“What we found here is correlation, not causation,” Shen was quick to say. “For the specific microbial alterations we discovered, they could potentially be part of the disease pathogenesis,” he said. “But the reverse is also possible: as the disease develops, it changes the physiology of the skin in ways that make it easier for certain microbes to grow. From this study alone, we can’t distinguish cause from consequence.”
Shen rehearsed some of the complexity. “AD itself is a common but complex disease … there are different contributing factors,” he said. “Genetics, microbial exposure, microbial species composition, and variation within microbial strains are all layered on top of one another,” he said. “We are starting to tease apart those layers, but we are not yet at a point where we can say how much each one contributes to disease.”
Replicating the findings in a different group is necessary. “This is a very targeted Australian population, so it could be interesting to see whether we can replicate some of these findings in independent cohorts, in a different population,” Shen said. Larger studies wouldn’t hurt, either. “There are 400 infants in the study, but for those two-to-three-month samples, we only get less than a hundred. It would be great if we can, in an independent study, expand that sample size to reassure that those are consistently reproducible markers to help with disease prediction,” he said.
Identifying trajectories of microbial transformation and providing mechanistic explanation would go beyond the simple correlations. “The next step would be to culture some of those microbes in the lab and test their characteristics and how they might be related to disease mechanisms,” Shen said.
Potentially, the results could lead to better anticipation of disease progression. “This is data that shows there’s definitely potential,” Shen said. “The next step would be really getting these observational findings into more translational settings—building machine learning models to see whether we can use those markers to help with disease prediction.” If specific microbes are shown to reliably predict disease, PCR could be used to easily quantify them in skin swab samples.
It is an open question what might be done for children with predicted increased risk of developing eczema and food allergies, but useful clues may exist in this dataset. One finding from the study was that Lactococcus cremoris levels decreased in AD + FA, and perhaps more striking, this was the only organism whose levels were inversely correlated with AD disease severity. Potentially, intervention could one day include topical probiotic formulations, as has been attempted in adult AD populations.
The group is also looking beyond the skin. “We have stool samples from the same babies; we’re actually trying to look at the so-called gut-skin axis,” Shen said. “In this published paper, we look at how skin microbiome could be associated with a skin disease; now we want to see how gut microbiome might be associated with FA and AD.”