Age-Specific Metabolomic Profiles in Children with Food Allergy

Abstract

Background.

Food allergy (FA) in young children is often associated with eczema, frequently directed to egg/cow milk allergens and has a higher chance of resolution, while FA that persists in older children has less chance of resolution and is less clearly associated with atopy.

Methods.

Children with FA (n=62) and healthy controls (n=28) were categorized into “younger” (≤5 years) and “older” (>5 years). Mass spectrometry-based untargeted metabolomic profiling as wells as cytokine profiling were performed on plasma samples in FA children in each age group.

Results.

Younger FA children manifested unique alterations in bile acids, polyamine metabolites and chemokines associated with Th2 responses, while older FA children displayed pronounced changes in long chain fatty acids, acylcarnitines and proinflammatory cytokines.

Conclusions.

FA children of different ages manifest unique metabolic changes which may reflect at least in part pathogenic mechanisms and environmental influences operative at different time points in the disease course.

Introduction

Food allergies (FA) represent a growing public health concern both in the US and worldwide [1, 2]. FA not only carry the risk of potentially life-threatening reactions, but they also significantly affect quality of life and are responsible for considerable healthcare costs [3, 4]. There is therefore an urgent need to achieve preventative and curative strategies that may be able to curb the current FA epidemic. Early introduction of allergenic foods in at-risk children has emerged as a promising approach to decrease the frequency of FA in young children [5, 6]. No preventative intervention is currently available to decrease the risk of FA in older children or adults, and the only available treatment for FA – namely oral desensitization - fails to foster permanent FA resolution in the majority of patients.

While FA in young children is often associated with eczema, more commonly directed towards chicken’s egg and cow milk and has a higher chance of resolution, FA that persists in older children is commonly elicited by a variety of food triggers, has a lower chance of resolution and is less consistently associated with other atopic manifestations [1]. These observations suggest the presence of different stages of FA that may results from differential pathogenic mechanisms and environmental influences operative at subsequent time points during the course of FA.

Untargeted metabolomic profiling is a powerful technique that has been extensively employed to dissect the pathogenesis of complex diseases, given its ability to detect alterations related to both genetic and/or environmental factors. To date, few studies have applied untargeted metabolomics to the study of FA in children, looking at established disease [7–11], severity [7], anaphylaxis [12], disease resolution [11] and food-specific profiles [13]. No study so far has directly compared metabolomic profiles of children of different age groups and disease duration.

Here, we compared untargeted metabolomic profiles of preschool and school-aged FA children with differential disease duration with those of age-matched healthy controls with the goal of identifying metabolic changes that are uniquely associated with the presence of FA at different ages, with the goal of gaining insights into causative mechanisms of FA at various disease stages. Results revealed shared and distinct metabolomic signatures associated with FA in younger versus older children, indicating both unique and overlapping underlying pathophysiological mechanisms operative in these age groups with differential duration of FA.

Methods

90 subjects ages 12 and younger were enrolled in this study, including 62 children with FA and 28 non atopic controls. Participants were recruited in the Allergy clinic at Boston Children’s Hospital and from the EASY (Environmental Assessment of Sleep in Youth) study, a NIH-funded study of the home environment and sleep outcomes in children (PI, Dr Phipatanakul). The study was approved by the Boston Children’s Hospital IRB. Informed consent was obtained from the children’s parents or legal guardians and children’s assent was obtained from age-appropriate participants.

FA diagnosis was made by an allergist during a clinic visit, based on the combination of positive IgE testing (skin testing and/or specific IgE) and recent history of symptoms consistent with IgE-mediated reaction (hives, angioedema, emesis, diarrhea, wheezing, hypotension etc.) developing within 2 hours of ingestion of the culprit food. Non-atopic controls had no history of IgE mediated conditions (asthma, atopic dermatitis, FA, allergic rhinitis). Exclusion criteria included history of non-IgE mediated or chronic inflammatory gastrointestinal disorders (i.e. eosinophilic esophagitis, inflammatory bowel disease etc.), use of immunosuppressants, and use of antibiotics within the previous 6 weeks.

Participants were further categorized based on age into subjects less or equal than five years (“FA young” and “control young”) and subjects older than five years (“FA old” and “control old”).

After collection, blood samples were spun, and serum stored at −80C.

Non-targeted global metabolomic profiles were generated through Metabolon Inc, research Triangle, NC, utilizing ultra-performance liquid chromatography/mass spectrometry. Metabolites were identified by their m/z retention time, and through a comparison to library entities of purified known standards.

Cytokine quantification was performed using Olink proximity extension assay (Target-48 panel; Waltham, MA). Analytes were measured from human plasma using an Olink Signature Q-100 instrument according to manufactirer’s protocol as described [14]. Olink NPX Signature software was used for data analysis and normalized proteinconcentrations were converted to absolute units (pg/ml) by calibration to internal standard curve.

Following log-transformation and imputation of missing values, if any, with the minimum observed value for each compound, each biochemical in the original scale was rescaled to set the median equal to 1 (scaled values). Pairwise comparisons were used to identify biochemicals that differed significantly between experimental groups. Student T-test, Welch T-test for unequal variances and Mann-Whitney tests were employed based on group size, metabolite distribution and proportion of missing/undetectable data. Metabolites with more than 1/3 of missing/undetectable data were excluded from the analyses. Regression analyses were performed to adjust for characteristics that showed statistically different distributions among study groups (atopic dermatitis and asthma). Random Forest Analysis was performed to provide an “importance” rank ordering of biochemicals that discriminate between groups. To observe data patterns between study groups, sparse partial least-squares discriminant analysis (sPLS-DA) was performed through MetaboAnalyst [15]. Five components were used to control the sparseness of the sPLS-DA model. Statistical analyses were carried out using Graphpad Prism version 7a, SPSS version 27 and Metaboanalyst.

Results

1. Study population characteristics

Population demographics and disease attributes are summarized in Table 1. A slightly higher, though not statistically significant, proportion of male participants was present in all groups except for younger controls. There was no significant difference in age between all FA children and all controls. The mean age was 2.5 years in the “FA young” group, 8.9 years in the “FA old” group, 3.6 years in the “control young” group, and 9.2 years in the “control old” group. While there was a significant age difference among the 4 groups - as directed by the study design -, no age difference was present in pairwise group comparisons (i.e. young FA vs young controls, and old FA vs old controls), allowing to detect FA-specific rather than age-dependent features.

Table 1.

Demographic characteristics of the study population.

	Food Allergy ≤ 5 years N=36	Food Allergy > 5 years N=26	Controls ≤ 5 years N=12	Controls > 5 years N=16
Age (years)*	2.5	8.9	3.6	9.2
Gender Male/Female	19/17	14/12	6/6	10/6
Ethnicity n (%)
White	16 (44.4%)	15 (57.7%)	6 (50.0%)	7 (43.8%)
- AA	7 (19.5%)	6 (23.1%)	3 (25.0%)	2 (12.5%)
- Hispanic	3 (8.3%)	1 (3.8%)	3 (25.0%)	5 (31.2%)
- Asian	7 (19.5%)	2 (7.7%)	0	0
- Other/Unknown	3 (8.3%)	2 (7.7%)	0	2 (12.5%)
Atopic dermatitis*	23/36 (63.9%)	9/26 (34.6%)	0/12	0/16
Asthma*	11/36 (30.6%)	14/26 (53.8%)	0/12	0/16
Allergic rhinitis	22/36 (61.1%)	21/26 (80.8%)	0/12	0/16

^*Significantly different distribution across study groups

With regard to FA characteristics, one quarter of enrolled children were diagnosed with one FA both within the young (9/36, 25.0%) and old (7/26, 26.9%) FA groups, while the majority of subjects suffered from two or more FA (Suppl. Fig 1A). The distribution of the most common FA triggers (cow milk, egg, peanut, tree nuts, sesame, wheat, fish/shellfish) among FA children by age is shown in Suppl. Fig. 1B. Young children had a higher prevalence of egg allergy, while fish and shellfish allergies were more common among older children. Disease duration (i.e. time from FA diagnosis to sample collection) was significantly different between the younger and older FA groups (1.5 vs. 6.2 years, respectively).

Atopic dermatitis was more frequent among young FA children, while a significantly larger proportion of old FA subjects carried a diagnosis of asthma (Table 1), consistent with the natural history of allergic disorders (so called “atopic march”). There was no difference in the rate of sensitization to environmental allergens between younger and older FA children in our cohort.

2. Metabolomic profiling

A total of 881 biochemicals were detected by metabolomic profiling in this study. After removal of xenobiotic, drug and chemical metabolites, as well as metabolites with large proportion of missing/undetectable values as described in the method section, 752 compounds of known identity were included in the analyses.

2.1. Comparison of FA children and controls independent of age

When comparing FA children of all ages to non-atopic controls, a total of 150 metabolites differed significantly between the two study groups (Suppl. Table 1). These include lipid metabolites (decreased levels of ceramides, sphingomyelins, lysophospholipids and increased levels bile acids and fatty acids in FA children), and amino acids (increased lysine, histidine, leucine, tyrosine and tryptophan metabolites among FA subjects). Of note, many of these metabolites have been previously reported to be differentially expressed in FA [7, 9]. A heatmap of the top 80 most dysregulated metabolites in FA children is shown in Suppl. Fig 2.

2.2. Comparison of FA children and controls by age groups

We then proceeded to compare the metabolomic profiles of age-matched groups of FA children and controls. A total of 110 metabolites were differentially expressed in young (≤ 5 years) FA children (Fig. 1, Suppl. Table 2), while 122 metabolites were significantly different between old (>5 years) FA children and controls (Fig. 2, Suppl. Table 3).

Figure 1.

Comparison of younger (≤5 years) FA children and age-matched controls. Panel A shows a heat-map of the top 80 metabolites significantly different (p<0.05) between the two groups. For each metabolite, a colorimetric representation of relative expression in each sample is shown according to the scale depicted on top. Panel B depicts pathway analysis showing the most commonly affected pathways and their directionality.

Figure 2.

Comparison of older (>5 years) FA children and age-matched controls. Panel A shows a heat-map of the top 80 metabolites significantly different (p<0.05) between the two groups. For each metabolite, a colorimetric representation of relative expression in each sample is shown according to the scale depicted on top. Panel B depicts pathway analysis showing the most commonly affected pathways and their directionality.

When looking at the comparison of young FA children and controls, 50.9 % of differentially expressed metabolites belong to lipid metabolic pathways, 24.5% to amino acids, 6.4% to nucleotides and 8.2% are cofactors/vitamins (Suppl Fig. 3A). The most common altered lipid pathways involved phospholipids (16.1%), short and medium chain fatty acids (16.1%), primary and secondary bile acids (12.5%), and sphingomyelins/ceramides (12.5%). Among amino acids, young FA children showed dysregulation of histidine (14.8%), methionine/cysteine (14.8%), polyamine (14.8%), arginine/proline (14.8%), and aromatic amino acids (11.1%). (Suppl Fig. 3 B). A heatmap of the most dysregulated metabolites in FA children younger than 5 years of age is shown in Fig. 1A. Fig. 1B shows the most significantly altered pathways as well as the proportion of up- or downregulated pathway components.

Among FA children ages 5 years and older, metabolomic alterations involve 65.6% of lipids, 18.0% of amino acids, 3.3% of nucleotides and 3.3% of cofactors/vitamins (Suppl Fig. 3A). Specifically, alterations in long chain fatty acids (16.4%), phospholipids (16.5%), acyl carnitine (13.9%), sphingomyelins/ceramides (13.9%) represent the most common lipid abnormalities, while the most common amino acid alterations involved the metabolism of aromatic amino acids (21.7%), leucine/isoleucine/valine (21.7%), histidine (17.4%) and arginine/proline (13%). (Suppl Fig. 3B). A heatmap of the 80 most dysregulated metabolites in old FA children is shown in Fig. 2A. Fig. 2B shows the most significantly altered pathways as well as well as the directionality and proportion of up- or downregulated pathway components.

In the direct comparison of young and old FA children, a total of 121 metabolites were differentially represented (Suppl. Table 4). Fig. 3A shows a sPLS-DA plot obtained from a supervised model for young vs old FA subjects. The weight of the first two components was 6.4% for the young and 7% for the old FA group. The clustering of subjects by age showed a significant tendency of separation between the two groups, with boundaries overlapping only slightly. Full separation was observed when comparing FA children and controls overall (Suppl. Fig. 4A), and by age group (Suppl. Fig. 4B, C). Fold change analysis identified 19 metabolites significantly downregulated, and 8 metabolites significantly upregulated in old as compared to young FA children (Fig. 3B). Fig. 3C shows the top 15 differentiating biochemicals by Random Forrest Analysis. The comparisons suggest key differences in glutamate (N-acetyl-aspartyl glutamate, pyroglutamine), histidine (imidazole lactate, N-acetylcarnosine), phenylalanine (phenylpyruvate, N-formylphenylalanine), bile acids (taurocholate, glycocholate), and polyamine (N1, N12-diacetylspermine) metabolisms.

Figure 3.

Comparison of younger (≤ 5 years) and older ( >5 years) FA children. Panel A shows a SPLS-DA plot of identification of FA subjects based on age group. Panel B shows a volcano plot of the most up- and downregulated metabolites in older FA children as compared to younger FA children. Panel C lists the top fifteen metabolites that most strongly discriminate between older and younger FA children as determined by Random Forrest Analysis.

Given the differential distribution of atopic dermatitis and asthma (Table 1), adjusted analyses were performed to account for the effect of atopic comorbidities. Of the 121 metabolites differentially expressed between the two FA age groups in unadjusted comparisons, 102/121 (84.3%) remained significantly different at the p<0.05 level following adjustment for the presence of asthma and eczema, with additional 14 retaining borderline significance (0.05<p<0.1) (Suppl. Table 4). These results suggest that the metabolic differences observed between younger and older children are uniquely driven by differential FA characteristics rather than co-occurring allergic conditions.

While overlapping metabolomic alterations were detected in the comparison of either young or old FA children and age matched controls, unique metabolic changes were also observed within each age-restricted group comparison.

FA children ≤ 5 years of age showed significant alterations in bile acids, in particular secondary bile acids (Suppl. Table 2), and polyamine metabolisms (Fig. 5), as well as lysine and methionine amino acids. Secondary bile acids derive from primary bile acids by the action of intestinal bacteria and, in turn, they modulate the composition of the gut microbiome [16, 17]. Secondary bile acids have been reported to be altered in FA children as compared to children with asthma, suggesting that bacterial enzymatic processing may be uniquely dysregulated in FA [7]. Lysine has been shown to be largely acetylated by the gut microbiome contributing to intestinal inflammatory disorders [18]. Polyamines, including putrescine, spermidine and spermine, also produced by the intestinal microbiota, regulate a variety of essential biological processes - such as cellular growth and differentiation [19, 20] - and have been implicated in multiple aspects of the innate and adaptive immune response [21–24]. Polyamine metabolism is schematically depicted in Fig. 4. Ornithine decarboxylase converts ornithine to putrescine, which in turn is metabolized to spermidine and spermine. Methionine can also enter polyamine metabolism through the action of methionine adenosyl transferase. Dysregulated polyamine metabolism has been linked to numerous human conditions [25, 26], especially cancer [27], though not previously to FA.

Figure 5.

Polyunsaturated fatty acid (PUFA) metabolism is uniquely dysregulated in food allergic children older than 5 years of age. PUFA metabolism is depicted together with pairwise comparisons of metabolites significantly different in older FA children and age-matched controls. Mean and SEM are shown.

Figure 4.

Polyamine metabolism is dysregulated in food allergic children ages 5 years and younger. Polyamine metabolism is depicted together with pairwise comparisons of metabolites significantly different in younger FA children and age-matched controls. Mean and SEM are shown.

FA children older than 5 years showed marked dysregulation in their lipid metabolic profile, with significant changes affecting acyl carnitine and medium- and long-chain fatty acid metabolisms, including long-chain polyunsaturated fatty acids (PUFAs) (Fig 3, Suppl. Table 3).

Acyl carnitines play important roles in cellular energy metabolism and have been implicated in the pathogenesis of many inflammatory disorders [28]. In one study looking at FA children, acyl carnitine levels were found to be lower at diagnosis as compared to controls [11]. PUFA metabolism is schematically depicted in Fig. 5. PUFAs encompass n-6 and n-3 fatty acids, which have pro- and anti-inflammatory properties respectively [29]. n-6 PUFAs have been previously associated with asthma both in children and adults [30–33]. Lower levels of n-3 PUFA have been reported in children with atopic dermatitis and allergic rhinitis [34], while higher levels have been associated with a decreased risk of allergy development [35]. In FA, n-3 PUFA have been implicated in both the sensitization and effector phase, suggesting a role for n-3 PUFA in decreasing allergy risk and preventing FA [36]. Lower levels of n-3 PUFA have been linked to the presence of FA and higher levels to FA resolution [11]. The dysregulation of PUFA we observed in old FA children with longer disease duration supports a role for this class of long-chain fatty acids in FA persistence beyond the first few years of life.

3. Validation cohort

We finally sought to validate our findings of key differences in altered metabolic pathways present in younger and older FA children by looking at a separate cohort of 70 children with FA, which was enrolled with similar recruitment criteria [7]. This cohort included 39 children ≤5 years and 31 children >5 years. When comparing FA children by age group and differential disease duration in this validation cohort, we again observed higher levels of polyamines (acetylspermidine, 4-acetamidobutanoate, N-acetyl-isoputreanine, N-acetyl-putrescine) and bile acids (glycol-alpha-muricholate, glycochenodeoxycholate, glycocholate) in younger children with FA, while older FA children displayed marked alterations in acyl carnitine and long chain fatty acid levels, including PUFAs. Suppl. Fig. 5A shows a heatmap of the 80 most dysregulated metabolites in the comparison of younger and older FA children in the validation cohort, while Suppl. Fig. 5B shows selected pair-wise comparisons. sPLS-DA plot obtained from a supervised model for young vs old FA subjects (Suppl. Fig. 4D). These replicated findings in an independent cohort add strong support to the notion that age specific metabolomic alterations are present in FA children which may be related to different disease stages and be reflective of pathogenic mechanisms operative at different time points throughout the disease course of FA.

4. Cytokine profiling

To further elucidate underlying mechanisms that distinguish the two patient groups, a panel of 48 cytokines was measured in plasma of 45 FA children (22 ≤ and 23 >5 years of age) and 22 controls (9 ≤ and 13 > 5 years of age). Three analytes (thymic stromal lymphopoietin/TSLP, interleukin (IL)-2, and IL-4) showed median levels below the limit of quantification for all groups and were removed from final analyses.

Independent of age, FA children displayed lower levels of C-X-C motif chemokine ligand 12 (CXCL12) and higher levels of C-C motif chemokine ligand 2 (CCL2), IL-27, IL-1b, IL-18, IL-17A, oxidized low density lipoprotein receptor 1 (OLR1), matrix metallopeptidase 12 (MMP12), CCL4, CCL7, CCL11, CCL13 and CCL19. When looking at age-restricted comparisons, FA children ≤ 5 years of age had lower levels of CXCL12 and increased levels of IL-18, IL-17C, CCL4 CCL13, CCL19, and MMP12 when compared to age-matched controls. Cytokine profile of FA children > 5 years of age was characterized by increased levels of IL-18, IL-6, OLR1 and oncostatin M (OSM) as compared to older controls (Fig. 6). In children ≤ 5 years of age we observed a strong correlation between the serum concentrations of CCL4, a chemokine implicated in allergic immune responses [37] and multiple polyamine metabolites, which were uniquely altered in this younger FA group (Suppl. Table 5). Among older FA children, multiple PUFA species were found to correlate strongly with the proinflammatory cytokines IL-27 and IL-6 (Suppl. Table 6), which have previously been shown to be associated with PUFA levels in a murine model of colitis [38].

Figure 6.

Cytokine alterations in FA children. Pairwise comparisons of cytokines significantly different in FA children independent of age (left), in FA children ≤ 5 years (center) and FA children > 5 years of age (right) as compared to non-atopic controls. Mean and SEM are shown.

Discussion:

In this study, untargeted metabolomic profiling of age-matched FA children and non-atopic controls revealed both shared and unique signatures associated with FA in children of different age groups and disease duration. In comparison to non-atopic controls, FA children ≤5 years manifested a prominent dysregulation of microbiome related metabolic pathways, including bile acids and polyamines while FA children >5 years displayed a signature of marked lipid alterations involving medium and long-chain fatty acids, including PUFA, and carnitine metabolism.

FA often starts early in life when it is strongly associated with the presence of atopic dermatitis. Food sensitization through an impaired skin barrier is thought to play a key role in the initiation of abnormal food directed IgE responses leading to FA, while food exposure through the oral route is thought to promote the development of tolerance through the engagement of gut mucosal immune responses in close interaction with the developing microbiome [39]. FA that develops in infancy is commonly directed towards foods such as cow milk and chicken egg and has a high chance of spontaneous resolution over time. FA that continues past school age is often directed towards a broader range of allergens that includes peanut, tree nuts, shellfish and tends to persist long-term with low chance of resolution. It appears reasonable to surmise that FA at different ages and duration may reflect differential and incompletely understood age- and stage-specific pathogenic mechanisms that are responsible for initiation vs persistence of disease. Metabolomic profiling was applied with the goal to unveil differences in mechanisms operative at different times of FA disease course.

The findings of marked alterations in bile acids, aromatic amino acid and – especially - polyamines metabolites in younger – but not older – FA children compared to age-matched, non-allergic controls suggest a prominent role for an altered microbiome in the early stages of FA, in line with multiple observations supporting a significant contribution of dysbiosis to FA, starting in infancy [40]. Secondary bile acids are polyamines are known to affect regulatory immune mechanisms [21, 41], including microbiome-imprinted RORgt₊ Treg cells [42, 43], suggesting a defect in regulatory mechanisms as key in FA development in early life as a result of dysbiosis, which may drive food-directed Th2 immune responses and expansion of the mast cell compartment in the gut. This was supported by the cytokine profile of younger food allergic children which was characterized by elevation of Th-2 associated cytokines such as CCL4, CCL13 and CCL19 [37, 44, 45], with strong correlation detected between polyamine metabolites and CCL4. In this younger FA group, microbial-directed therapies [46] could be particularly effective in affecting local gut homeostasis and possibly preventing FA. Of note, polyamines have not previously been reported to be altered in FA children and represent candidate biomarkers for early FA development.

In older FA children, the continued dysregulation of aromatic amino acid metabolites in comparison to age matched controls suggests an ongoing contribution of dysbiosis to all stages of FA. However, marked changes in lipid metabolism uniquely observed in older children may indicate additional altered immune mechanisms leading to the continued presence of FA. Dysregulated lipids included PUFAs and acylcarnitines which were elevated in older FA children as compared to both non-atopic controls and to younger FA children. PUFAs have been previously implicated in multiple steps of FA pathogenesis [36]. Specific PUFAs have been shown to modulate the ability of dendritic cells to induce the production of Th2 cytokines in vitro, suggesting a role in perpetuating allergic responses [47]. Additionally, arachidonic acid, a n-6 PUFA, is a precursor to many inflammatory mediators involved in the allergic immune response. Our observations that cytokines and proteins with proinflammatory function such as IL-6, IL-18, OSM and OLR1 [48, 49] are increased in older FA children and that PUFA levels strongly correlate with IL-27 and IL-6 levels suggest that their disease course may be characterized by evolution towards a proinflammatory state that promotes disease persistence.

Children in the older group had a significantly longer disease duration as compared to younger children (6.2 vs 1.5 years, respectively) suggesting that findings in the older group may be related to mechanisms of FA persistence, rather than late onset. The current understanding of the factors associated with FA persistence is rather limited, which is reflected in the lack of available prognostic biomarkers. A recent study identified secondary bile acids of the alternative pathway as candidate biomarkers of FA persistence, possibly in relation to microbiome alterations [50]. Another study has linked n-3 PUFA levels to FA prognosis [11]. Here we identify prominent lipid alterations present in older children with longer disease duration, further suggesting a possible role for lipid mediators as candidate biomarkers and mechanistic players in FA persistence.

A strength in our study involves the recruitment of well phenotyped subjects with FA diagnosis of FA made by an allergy physician, an approach that reduces the bias of self-reported FA. The availability of detailed information regarding concomitant allergic diseases allowed us to perform adjusted analyses taking into consideration the effect of overlapping allergic comorbidities, which confirmed that differences observed between groups of younger and older children were uniquely driven by FA. Finally, the availability of an independent cohort of FA children allowed us to validate our findings in a separate patient population, suggesting that age specific differences are indeed present in FA children and likely linked to disease stages.

Limitations of our study include a small sample size which caused insufficient power to reach significance when adjusting for multiple comparisons. A focus on pathways with multiple dysregulated components - rather than on individual metabolites - and replication in a separate population were sought to strengthen the validity of our observations despite the size and power limitations. Children of different ages may have different dietary habits, both in relation to their FA and to age related food behaviors. Our study included children mostly 1 year of age and older, after the introduction of solid foods. In addition, the distribution of FA between younger and older FA children was similar, minimizing the effects of individual food avoidances. Still, larger studies that include dietary recall may be needed in the future to fully address the potential impact of dietary restrictions and food-related habits on metabolomic profiles. Finally, the cross-sectional nature of our study does not allow to fully assess if and how individual metabolomic profiles change within the same individuals over time as FA persist. Longitudinal studies that follow FA patients along their disease trajectory will be key in addressing this limitation and in allowing to uncover prognostic biomarkers of FA.

Conclusion:

In summary, our study revealed unique metabolomic changes present in children of different ages and FA duration, suggesting that differential mechanisms may drive FA in early stages, when a greater chance of resolution is present, while additional mechanisms may become operative later in life and impair FA resolution causing disease persistence. Our observations suggest a role for metabolomic profiling in dissecting FA endotypes and identifying candidate biomarkers and strongly points to age- and other phenotype-specific investigations as necessary to drive personalized medicine in the care of patients with FA.

Supplementary Material

1

Supplementary Table 1. List of metabolites significantly different between FA children and controls. Metabolite name, pathway, p-value and FDR are listed.

Supplementary Table 2. List of metabolites significantly different between younger FA children (≤5 years) and age-matched controls. Metabolite name, pathway, p-value and FDR are listed.

Supplementary Table 3. List of metabolites significantly different between older FA children (>5 years) and age-matched controls. Metabolite name, pathway, p-value and FDR are listed.

Supplementary Table 4. List of metabolites significantly different between younger (≤5 years) and older (>5 years) FA children. Metabolite name, pathway, unadjusted p-value and p-value adjusted for asthma and eczema are listed.

Supplementary Table 5. Correlation between plasma cytokines and polyamine metabolite levels in FA children ≤5 years of age.

Supplementary Table 6. Correlation between plasma cytokines and PUFA levels in FA children >5 years of age.

2

Supplementary Figure 1. FA characteristics by age group. Panel A shows the number of FA (one, two or three and more). Panel B shows the proportions of specific food allergies.

Supplementary Figure 2. Heat-map of metabolites significantly different (p<0.05) FA children and non-atopic controls. For each metabolite, a colorimetric representation of relative expression in each sample is shown according to the scale depicted on top.

Supplementary Figure 3. Younger (≤ 5 years) and Older (>5 years) children with FA display differential alterations in overall metabolic pathways (A) as well as in specific aminoacid and lipid metabolites (B) when compared to age-matched controls.

Supplementary Figure 4. SPLS-DA identification plots of FA subjects of all ages vs controls (A), of younger (≤ 5 years), B) vs age matched controls, and older (>5 year, C) FA children vs age matched controls in the study population. Panel D displays the SPLS-DA identification plot of FA subjects based on age group in the validation cohort.

Supplementary Figure 5. Comparison of younger (≤5 years) and older (>5 years) FA children in confirmatory cohort. Panel A shows a heat-map of the top 80 metabolites significantly different (p<0.05) between the two groups. Panel B shows pairwise comparisons of metabolites significantly different in younger vs older FA children in confirmatory cohort.

Highlights:

• Distinct metabolomic profiles characterize recently established food allergy in young children as compared to persistent food allergy in older children, suggesting differential mechanisms underlying food allergy stages.

• Food allergy in young children is associated with marked alteration of microbiome-related metabolites, including bile acids and polyamines

• Food allergy in older children is associated with marked alteration of lipid mediators, including PUFAs and acyl carnitines.

• Polyamines are uniquely altered in early stages of food allergy and represent a novel class of candidate food allergy biomarkers