Longitudinal dynamics of the gut microbiome and metabolome in peanut allergy development

Abstract

Background:

Rising rates of peanut allergy motivate investigations of its development to inform prevention and therapy. Microbiota and the metabolites they produce shape food allergy risk.

Objective:

To gain insight into gut microbiome and metabolome dynamics in the development of peanut allergy.

Methods:

We performed a longitudinal, integrative study of the gut microbiome and metabolome of infants with allergy risk factors but no peanut allergy from a multi-center cohort who were followed through mid-childhood. We performed 16S rRNA sequencing, short chain fatty acid measurements, and global metabolome profiling of fecal samples at infancy and at mid-childhood.

Results:

In this longitudinal, multi-center sample (n=122), 28.7% of infants developed peanut allergy by mid-childhood (mean age 9 years). Lower infant gut microbiome diversity was associated with peanut allergy development (P=0.014). Temporal changes in the relative abundance of specific microbiota and gut metabolite levels significantly differed in children who developed peanut allergy. Peanut allergy-bound children had different abundance trajectories of Clostridium sensu stricto 1 sp. (false discovery rate (FDR)=0.015) and Bifidobacterium sp. (FDR=0.033), with butyrate (FDR=0.045) and isovalerate (FDR=0.036) decreasing over time. Metabolites associated with peanut allergy development clustered within the histidine metabolism pathway. Positive correlations between microbiota, butyrate, and isovalerate and negative correlations with histamine marked the peanut allergy-free network.

Conclusion:

The temporal dynamics of the gut microbiome and metabolome in early childhood are distinct for children who develop peanut allergy. These findings inform our thinking on the mechanisms underlying and strategies for potentially preventing peanut allergy.

Capsule summary:

This longitudinal multi-center study of children identified temporal changes between infancy and mid-childhood in gut microbiota and metabolite levels associated with peanut allergy development.

Graphical Abstract

Introduction

Food allergy is a growing health concern.¹ Among food allergies in children, peanut allergy is one of the most common.^{2, 3} It is the most likely food allergy to cause life-threatening anaphylaxis and is rarely outgrown, with 80% of peanut allergy persisting into adulthood.⁴ There is no cure for peanut allergy, and the condition severely impacts quality of life.⁵ Given rising rates of peanut allergy, there has been much interest in learning about its development to inform potential prevention and therapy.

Increasing numbers of studies suggest that gut microbiota play a role in food allergy.^6–10 The gut microbiome in early life is thought to be especially important given its rapid evolution during this period and its influential cross-talk with the developing immune system.^{6, 11} The gut microbiome stabilizes with age, becoming resilient in adulthood in the absence of extreme stressors.¹² Multiple studies have found that gut microbial composition differs in individuals with and without food allergy,^6–8 and distinct microbial signatures for food allergy to different foods have also been reported.⁸ Further, gut microbiota are distinct in food allergic infants whose food allergy will persist vs. resolve in later childhood.¹³ The mechanisms by which gut microbiota shape host immune function and metabolism are multi-faceted, but the metabolites they produce, such as short chain fatty acids and others, are thought to be primary mediators.^{6, 7, 14, 15}

To gain insight into gut microbiome and metabolome dynamics in the development of peanut allergy, we performed this longitudinal study of a multi-center early-life cohort of 122 children at risk for developing peanut allergy (Figure 1). Examination of the children’s gut microbiomes and metabolomes at infancy when all were peanut allergy-free and later at mid-childhood after participants did or did not develop peanut allergy enabled us to identify changes in gut microbiota and metabolites associated with the development of peanut allergy.

Figure 1. Overview of the study and analytic flow.

(A) Stool samples from 122 children were collected at two time points (infancy and mid-childhood) for gut microbiome and metabolome profiling. All children had no peanut allergy (NPA) at infancy but had risk factors for peanut allergy (PA) development. 35 developed peanut allergy by mid-childhood, and 87 remained free of peanut allergy. (B) The analytic flow included the following steps: Alpha diversity and species abundance of the gut microbiome were analyzed at infancy ①, mid-childhood ②, and longitudinally ③. Gut metabolite levels were also analyzed at infancy ④, mid-childhood ⑤, and longitudinally ⑥. Integrated networks were then built to assess relationships between temporal changes in gut microbiota and metabolite levels associated with the development of peanut allergy ⑦.

Methods

Participants, sample collection, and definition of peanut allergy

Participants of this study included 122 children from the multi-center NIAID Consortium for Food Allergy Research (CoFAR) Observational Study (CoFAR2) who provided stool samples at both infancy and mid-childhood.¹⁶ The recruitment and clinical characteristics of these CoFAR2 subjects have been previously described.¹⁶ Briefly, 511 children were recruited at age 3 to 15 months from five US sites (New York, NY, Baltimore, MD, Little Rock, AR, Denver, CO, Durham, NC) with study visits occurring between 2006 and 2015.¹⁶ The cohort was designed as a longitudinal study of infants at high risk for developing peanut allergy, and inclusion criteria included egg allergy, milk allergy, and/or moderate to severe atopic dermatitis with a positive egg and/or milk skin prick test at enrollment, but no known peanut allergy.¹⁶ Egg and milk allergy were based on allergist assessment of convincing history of allergic reaction to milk and/or egg with a positive skin prick test to the triggering food. Clinical phenotyping of the subjects, including assessments of peanut allergy, egg allergy, milk allergy, and atopic dermatitis, were performed at 6–12 month intervals between enrollment at infancy and mid-childhood (mean age 9 years, SD 0.6 years). Sex was determined based on parental report. All subjects were instructed to provide stool samples at baseline and were invited to submit a follow up sample at mid-childhood. Stool samples were collected from 491 of the children at baseline and from 122 of the children at mid-childhood. Samples were immediately stored at −80°C upon receipt.

Children were categorized as PA if they developed peanut allergy by mid-childhood and not peanut allergic (NPA) if they did not develop peanut allergy by mid-childhood. For this study, peanut allergy was defined based on: (1) confirmed IgE-mediated reaction (e.g. positive doctor-supervised oral food challenge to peanut and sensitization to peanut), or (2) convincing IgE-mediated reaction (e.g. convincing reaction and sensitization to peanut) at any visit.¹⁶ This was “alternate definition 1” of the parent CoFAR2 study¹⁶. We used this definition rather than the main definition to avoid inclusion of less convincing peanut allergy in case ascertainment, as the main definition¹⁶ also included those with peanut sensitization but no history of reaction.

Microbiome profiling

All stool samples from the infancy and mid-childhood time points underwent DNA isolation, library preparation, and sequencing in a single batch to minimize technical effects due to batch. DNA extraction from the stool samples was performed using the DNeasy PowerLyzer PowerSoil Kit (QIAGEN, Hilden, Germany). 16S rRNA library preparation was performed according to the Illumina MiSeq 16S Metagenomic Sequencing Library Preparation protocol. Both agarose DNA gel electrophoresis and Bioanalyzer were used to assess and quantify final PCR products. Dual indices and Illumina sequencing adapters were attached to the V3-V4 amplicon of 16S rRNA by using the Nextera XT Index Kit, v2 (sets A, B, and C; Illumina, San Diego, CA). PCR blanks were included with the samples for sequencing as negative controls. Sequencing was performed on amplified 16S rRNA by using the Illumina MiSeq V2 kit with 2 × 250 bp paired-end reads. Sequences were denoised and quality controlled using DADA2¹⁷ plugin in Qiime2¹⁸. The sequences were summarized to the amplicon sequence variant (ASV) table via DADA2¹⁷, and nonzero counts detected from the PCR blanks were subtracted from all other samples. The Silva 138 database¹⁹ was used to generate a phylogenetic tree and classify taxonomy of the reads. All samples met the minimum sequences per sample threshold of 2,000. Alpha diversity was assessed by Simpson’s E index for evenness and Simpson’s diversity index²⁰ using Qiime2.¹⁸ Species with zero count in 2/3 or more of samples and relative abundance less than 0.5% were removed prior to analyses.

SCFA profiling

Stool samples were weighed into bead blaster tubes with zircon beads and homogenized in 100% methanol to a final concentration of 10mg/mL including deuterated SCFA internal standards (CDN Isotopes).²¹ A pooled sample homogenate was created and aliquotted to serve as control samples. Control blanks were generated by transferring 1mL of methanol (instrument blank). Aliquoted pooled sample controls were then subjected to three successive rounds of extraction to deplete SCFAs but retain the insoluble particles to generate null matrix material. A standard curve for the cocktail of SCFAs was prepared in null matrix, and these samples were extracted side-by-side with the study samples and sequence blanks (either methanol or null matrix). The resulting SCFA extracts were then derivatized as described²² for liquid chromatography tandem mass spectrometry (LCMS) with methanol being used instead of acetonitrile. Samples were analyzed with a QExactive HF coupled to a Dionex Ultimate 3000. A Waters BEH C18 column (2.1 × 150mm, 1.7μm) was used with Buffer A as 0.1% formic acid in water, and Buffer B as 0.1% formic acid in acetonitrile. A gradient elution was used (15%B to 55%B in 9 min, 200μL/min) and the mass spectrometer was operated in negative ion mode (HESI, −3.5kV). The order of acquisition was randomized and several control blocks including blanks and external standards were interspersed throughout the run to assess instrument performance and quality control of the derivatization protocol. All analytes were corrected to their respective internal standard, and the resulting ratio was interpolated against the matrix-controlled standard curve to quantify the level of SCFA in each sample. The limit of detection was determined from null-matrix samples for each analyte, and this limit was imputed for any point falling below the limit of detection. The interpolated values were used for the downstream analyses, and the imputed ratio threshold values were used for samples below the limit of detection.

Global metabolome profiling

Global metabolome profiling was performed for 40 samples (both infant and mid-childhood samples from 10 NPA and 10 PA children randomly selected). Fecal samples were processed with a polar metabolite extraction, scaling the extraction to a ratio of 10mg/mL sample/extraction solvent. The resulting polar metabolite extracts for each sample were analyzed with the global polar LCMS platform. To identify putative molecules in the samples, the available MS/MS spectra from the data-dependent acquisition were searched against a data analysis pipeline including both the NIST17MS/MS and METLIN spectral libraries. Across all samples 64,482 tandem MS (MS/MS) spectra were matched against the spectral libraries, and then refined into a list of 2,436 putatively identified compounds (RevDot >900, Unique InChiKey). These hits were then quantified in a relative fashion based on the theoretical m/z of the identified compound as the putative ion type e.g., [M+H]+, at the consensus retention time. The list of putative compounds was further reduced to 2,189 by excluding any hits that resulted in the same metabolite name after identifier conversion (e.g., Glutamic acid, L-Glutamate), retaining the higher intensity row as applicable, and the resulting data were subjected to a 3X signal-to-noise threshold. Overall 1,825 compounds were detected in at least 2 samples with 1,008 being detected in at least 20 samples, and 177 metabolites being detected in all 40 samples. Global metabolites with mean intensity less than 1,000 were further excluded, yielding 1,967 metabolites for analysis.

Statistical tests and covariates

Alpha diversity metrics (Simpson’s E and Simpson’s diversity index), the relative abundances of species, SCFA levels, and global metabolite levels were log-transformed prior to model building. Cross-sectional differences in these measures between the NPA and PA groups at infancy and at mid-childhood were analyzed using linear regression models adjusted for participants’ milk allergy, egg allergy and atopic dermatitis. As age, sex, and race/ethnicity did not differ between the NPA and PA groups within infants or within children at mid-childhood (Table 1), we determined that they were unlikely to be potential confounders of cross-sectional analyses respectively limited to the infancy time point and the mid-childhood time point.

Table 1:

Characteristics of the children studied

	All (n=122)	No Peanut Allergy (NPA) (n=87)	Peanut Allergy (PA) (n=35)	p-value
Age
Infant, months	10.4 (3.1)	10.5 (3.2)	10.0 (2.9)	0.38
Mid-childhood, years	9.0 (0.6)	9.1 (0.5)	8.9 (0.6)	0.23
Sex (female)	38 (31%)	27 (31%)	11 (31%)	1.00
Race/ethnicity				0.16
White	92 (75%)	61 (70%)	31 (89%)
Black	13 (11%)	12 (14%)	1 (3%)
Latinx	4 (3%)	3 (3%)	1 (3%)
Other	13 (11%)	11 (13%)	2 (6%)
Peanut sIgE (kUA/L)
Infant	12.51 (24.54)	12.97 (26.05)	11.36 (20.46)	0.72
Mid-childhood	25.05 (36.78)	25.49 (36.57)	23.91 (37.87)	0.84
Peanut SPT (wheal mm)
Infant	5.2 (5.3)	4.8 (5.2)	6.2 (5.5)	0.20
Mid-childhood	9.7 (8.3)	8.8 (8.6)	11.9 (7.2)	0.048
Atopic dermatitis				0.63
Infant
None	7 (6%)	6 (7%)	1 (3%)
Mild	19 (16%)	12 (14%)	7 (20%)
Moderate	64 (52%)	44 (51%)	20 (57%)
Severe	30 (25%)	23 (26%)	7 (20%)
Mid-childhood				0.52
None	46 (38%)	36 (41%)	10 (29%)
Mild	47 (39%)	32 (37%)	15 (43%)
Moderate	25 (20%)	18 (21%)	7 (20%)
Severe	2 (2%)	1 (1%)	1 (3%)
Milk allergic
Infant	57 (47%)	38 (44%)	19 (54%)	0.32
Mid-childhood	30 (25%)	27 (31%)	3 (9%)	0.017
Egg allergic
Infant	36 (30%)	25 (29%)	11 (31%)	0.83
Mid-childhood	25 (21%)	21 (24%)	4 (11%)	0.21
Antibiotics
Infant (any during lifetime)	78 (64%)	57 (66%)	21 (60%)	0.68
Mid-childhood (since last visit)	36 (30%)	22 (25%)	14 (42%)	0.078
Breast fed - Infant				0.72
Never	14 (11%)	11 (13%)	3 (9%)
Yes - currently	46 (38%)	31 (36%)	15 (43%)
Yes - but no longer	62 (51%)	45 (52%)	17 (49%)
Mode of delivery (vaginal)	78 (64%)	54 (62%)	24 (69%)	0.54
Solid food introduced - Infant	115 (94%)	81 (93%)	34 (97%)	0.67

^*Fisher Exact test for categorical variables, two sided t-test for continuous variables.

Mean (SD) or number (%) are shown.

For the longitudinal analyses, linear mixed effect (LME) models were built with each of the measures as the dependent variable and the following independent variables: age, peanut allergy, egg allergy, milk allergy, and atopic dermatitis as fixed effect variables and subject as a random effect. Second models for each measure included an additional interaction term between time and peanut allergy. For each measure, a likelihood ratio test (LRT) was then performed to compare the base model to the second model. R package lme4²³ was used for LME modeling and R package lmtest²⁴ was used for LRT.

For all analyses of ASV, SCFA, and global metabolite data, we adjusted for multiple testing using a permutation-based approach with 10,000 iterations to calculate false discovery rate (FDR) values.^{25, 26}

Global metabolite pathway analyses

Pathway analysis using MetaboAnalyst 5.0²⁷ was performed on the 139 global metabolites identified as significantly associated (FDR ≤ 0.05) with the development of peanut allergy from the longitudinal analysis. The hypergeometric test was used for enrichment calculations, and relative-betweenness centrality was used for topology analysis²⁷. The topology analysis considered pathway structure and calculated impact scores for metabolites depending in their connectivity vs. isolation²⁷.

Integrated network construction

Change in relative abundance and change in metabolite levels between infancy and mid-childhood were used for the network construction. All species data were used except for species with zero count in 2/3 or more of samples and relative abundance less than 0.5%. All SCFA data were used and global metabolites were filtered to include the 139 metabolites associated with the development of peanut allergy from the longitudinal analysis (FDR ≤ 0.05). The delta values were log transformed and scaled from 0 to 1. Separate networks for NPA and PA children were built using the xMWAS algorithm²⁸, where sparse partial least squares is used for network integration and an approximation of Pearson correlation is performed for edge calculation. Edges were filtered to P≤0.05 and correlation r≥0.3. Because not all subjects underwent global metabolome profiling, separate microbiome-SCFA and microbiome-global metabolite networks were built to map the topologies of the PA and NPA and then merged to form the final networks for PA and NPA. Communities within the networks were identified by the xMWAS algorithm.²⁸

Results

Characteristics of the study cohort

The study sample included 122 children from the multi-center NIAID Consortium for Food Allergy Research Observational Study (CoFAR2) who provided stool samples at infancy and mid-childhood.¹⁶ The characteristics of these 122 children were similar to those of the overall cohort (Table E1). Briefly, the CoFAR2 cohort included 511 children who were recruited at age 3 to 15 months from five US sites and followed until mid-childhood (mean age 9 years, SD 0.6 years).¹⁶ The cohort was designed to follow infants who had no peanut allergy but had risk factors for developing peanut allergy at 6–12 month intervals through mid-childhood. Inclusion criteria included likely egg allergy, milk allergy, and/or moderate to severe atopic dermatitis with a positive egg and/or milk skin prick test at enrollment, but no known peanut allergy.¹⁶ All subjects were instructed to provide stool samples at baseline and invited to submit a follow up sample at mid-childhood (median age 9 years, interquartile range 0.73 years).

Among these 122 children, 35 (28.7%) developed peanut allergy by mid-childhood (Figure 1A, Table 1), defined based on confirmed or convincing IgE-mediated reaction.¹⁶ The assessments of peanut allergy were by made by CoFAR allergist/immunologist clinician investigators according to CoFAR guidelines.¹⁶ There were no significant differences in the characteristics of the 35 infants who later developed peanut allergy (PA) vs. the 87 infants who did not develop peanut allergy (no peanut allergy, NPA), although the peanut allergy-bound group trended toward larger peanut skin prick test (SPT) wheal size at infancy (Table 1). Children who had never experienced an allergic reaction to peanut but were sensitized to peanut were classified as NPA, likely contributing to higher than expected peanut specific IgE and SPT measures in the NPA group. At mid-childhood, the PA children had significantly larger peanut SPT wheals (mean 11.9mm vs. 8.8mm, P=0.048) and lower prevalence of milk allergy (9% vs. 31%, P=0.017) compared to NPA children (Table 1).

To study how gut microbiota and metabolite levels at infancy, mid-childhood, and over time might be associated with the development of peanut allergy, we assayed the infant and mid-childhood stool samples from these 122 children by 16S rRNA sequencing and measurement of metabolite levels. After quality control of the 16S rRNA data generated, 4,647 amplicon sequence variants (ASVs) with mean frequency of 25,263 remained for analyses. Short chain fatty acid (SCFA) levels were measured for all subjects at infancy and mid-childhood. Because global metabolome profiling is resource-intensive, global metabolome was measured for a subset of the samples (n=40, including both infant and mid-childhood stool samples from 10 NPA and 10 PA children randomly selected). After QC of the global metabolome data, 1,967 global metabolites remained for the downstream analyses.

Diversity of the gut microbiome is lower in children who later develop peanut allergy

A cross-sectional analysis comparing gut microbiome alpha diversity metrics at infancy revealed lower infant gut microbiome evenness as measured by Simpson’s E index among infants who later developed peanut allergy by mid-childhood compared to infants who remained free of peanut allergy (P=0.014, Figure 2A). At mid-childhood, however, both groups had comparable evenness with no significant difference (Figure 2B). Among infants who later developed peanut allergy, microbiome evenness was low in infancy and increased over time until mid-childhood, while NPA children demonstrated a steady level of evenness (Figures 2C). Longitudinal analysis revealed significant change in evenness over time associated with the development of peanut allergy (P=0.015). To additionally assess richness, Simpson’s diversity index was also calculated and showed no significant differences between groups at infancy, mid-childhood, or over time. Sensitivity models adding race/ethnicity, asthma, and number of allergen sensitizations as additional covariates did not change the significance of the results (Table E2). An additional sensitivity analysis comparing children eating peanut to children sensitized but without history of peanut ingestion showed no difference in Simpson’s E index at infancy or at mid-childhood (Figure E1),

Figure 2. Diversity of the gut microbiome at infancy, mid-childhood, and over time in a cohort of 122 children.

(A) Species evenness at infancy by Simpson’s E index in infants who later developed peanut allergy (PA, n=35) and infants who remained free of peanut allergy (no peanut allergy, NPA, n=87). Median and IQR are indicated by the box plots. P-value is from linear regression adjusted for participants’ milk allergy, egg allergy, and atopic dermatitis status. (B) Species evenness at mid-childhood for children who did (PA, n=35) and did not (NPA, n=87) develop peanut allergy. Median and IQR are indicated by the box plots. P-value is from linear regression adjusted for participants’ milk allergy, egg allergy, and atopic dermatitis status. (C) Change in species evenness over time was associated with the development of peanut allergy. Mean and standard error at infancy and mid-childhood are shown for children who did (PA, n=35) and did not (NPA, n=87) develop peanut allergy. P-value is from likelihood ratio test of linear mixed effects models adjusted for participants’ milk allergy, egg allergy, and atopic dermatitis status.

Changes in the relative abundances of gut microbial species over time are associated with the development of peanut allergy

At infancy, the relative abundance of Clostridium sensu stricto 1 sp. was higher in infants who remained free from peanut allergy through mid-childhood compared to infants who later developed peanut allergy, while Streptococcus sp. showed the opposite pattern (Figure 3A). We additionally tested this finding in the larger sample of 491 CoFAR2 infants with stool samples. Consistent with the results shown in Figure 3A, the relative abundance of Clostridium sp. was higher in NPA vs. PA (β coefficient = −0.11, FDR = 0.015), and the relative abundance of Streptococcus sp. (β coefficient = 0.091, FDR = 0.047) was higher in PA vs. NPA in the larger sample. At mid-childhood, children without peanut allergy had higher relative abundance of Bifidobacterium sp. than children with peanut allergy (Figure 3B). Longitudinal analyses of the infant and mid-childhood data showed that changes in the relative abundances of Clostridium sensu stricto 1 sp. and Bifidobacterium sp. over time were associated with the development of peanut allergy (Figure 3C). The two species showed opposite directions of temporal change between the children who did and did not develop peanut allergy. Clostridium sensu stricto 1 sp. was initially low at infancy in children who later developed peanut allergy and then high at mid-childhood compared to children who remained peanut allergy-free, while Bifidobacterium sp. showed the opposite dynamics. Although Streptococcus sp. was higher among infants who later developed peanut allergy (Figure 3A), longitudinal modeling did not show statistically significant change in its relative abundance over time associated with peanut allergy development (Figure 3C).

Figure 3.

Gut microbial species associated with the development of peanut allergy in a cohort of 122 children, of whom 35 developed peanut allergy (PA) and 87 remained without peanut allergy (NPA) by mid-childhood. (A) At infancy, the relative abundances of Clostridium sensu stricto 1 sp. and Streptococcus sp. were significantly associated (FDR ≤ 0.05) with the later development of peanut allergy. Beta coefficients are shown from linear regression models adjusted for milk allergy, egg allergy and atopic dermatitis. (B) At mid-childhood, higher relative abundance of Bifidobacterium sp. was associated (FDR ≤ 0.05) with no peanut allergy. Beta coefficient is shown from linear regression models adjusted for milk allergy, egg allergy and atopic dermatitis. (C) Change in the relative abundances of Clostridium sensu stricto 1 sp. and Bifidobacterium sp. over time were associated with peanut allergy development. Mean and standard error at infancy and mid-childhood are shown. FDR is from the likelihood ratio test of linear mixed effects models. Models were adjusted for participants’ milk allergy, egg allergy, and atopic dermatitis status. For comparison, results for Streptococcus sp. are also shown.

Temporal changes in fecal butyrate and isovalerate levels are associated with peanut allergy development

To complement our examination of the gut microbiome of these children, we also measured fecal levels of six SCFAs at infancy and mid-childhood: acetate, butyrate, propionate, isovalerate, isobutyrate, and valerate. Acetate, propionate, and butyrate had the highest detectable levels across all children at both time points (Figure 4A, 4B). Cross-sectional comparisons of the SCFA levels between the children who did and did not develop peanut allergy showed no significant differences for any SCFA at infancy (Figure 4A) or at mid-childhood (Figure 4B). Among the infants who developed peanut allergy by mid-childhood, butyrate and isovalerate levels significantly decreased during the interval, while infants who remained peanut allergy-free showed increased levels for butyrate and steady levels for isovalerate (Figure 4C). Linear mixed effects models revealed significant temporal changes in butyrate and isovalerate levels between infancy and mid-childhood associated with peanut allergy development (Figure 4C).

Figure 4.

Gut short chain fatty acid (SCFA) levels and the development of peanut allergy in a cohort of 122 children (NPA n=87, PA n=35). (A) SCFA levels at infancy, stratified by peanut allergy status at mid-childhood. Linear regression models adjusted for milk allergy, egg allergy, and atopic dermatitis showed no significant difference (FDR>0.05) in gut SCFA levels between infants who did (PA) and did not (NPA) later develop peanut allergy. Median and IQR are indicated by the box plots. (B) SCFA levels at mid-childhood, stratified by peanut allergy status at mid-childhood. Linear regression models adjusted for milk allergy, egg allergy and atopic dermatitis showed no significant difference (FDR>0.05) in gut SCFA levels between mid-childhood participants with and without peanut allergy. Median and IQR are indicated by the box plots. (C) Gut SCFAs with significant (FDR≤0.05) change over time associated with the development of peanut allergy. Mean and standard error at infancy and mid-childhood are shown. FDR is from the likelihood ratio test of linear mixed effects models adjusted for participants’ milk allergy, egg allergy, and atopic dermatitis status.

Changes in global metabolites within the histidine metabolism pathway are associated with peanut allergy development.

As a secondary analysis to further examine gut metabolites in the development of peanut allergy, we additionally assayed the global metabolome at infancy and mid-childhood in a subset of the study sample (40 samples, including paired infant and mid-childhood samples from 10 PA and 10 NPA children randomly selected). Among 1,967 putatively identified global metabolites, changes in the levels of 139 metabolites between infancy and mid-childhood were significantly associated (FDR ≤ 0.05) with peanut allergy development (Table E3). Pathway analysis using MetaboAnalyst²⁷ revealed that these global metabolites were significantly associated with a single pathway for histidine metabolism (FDR = 0.037, pathway impact = 0.28) (Figure 5).

Figure 5.

Pathway analysis of global metabolites associated with the development of peanut allergy (n=20 including 10 NPA and 10 PA participants) Linear mixed effect models adjusted for participants’ egg allergy, milk allergy, and atopic dermatitis status identified 139 global metabolites with significant (FDR≤0.05) changes over time associated with the development of peanut allergy. Pathway analysis of these 139 metabolites revealed a single pathway, histidine metabolism, with FDR≤0.05. The dotted line indicates FDR=0.05. Larger node size indicates higher pathway impact, a combination of the centrality and pathway enrichment calculated by MetaboAnalyst⁵¹.

Integrated network analysis reveals distinct microbiota–metabolite relationships among children who did and did not develop peanut allergy

To integrate our findings of gut microbiota and gut metabolites associated with development of peanut allergy, we next constructed integrated networks to characterize associations between temporal changes in the relative abundances of gut microbial species and metabolite levels. Using the xMWAS algorithm²⁸ for integrated network construction, we built separate networks for children who remained free of peanut allergy (Figure 6A) and those who developed peanut allergy (Figure 6B) that revealed distinct microbiota-metabolite relationships for the two groups. Among children who remained free of peanut allergy, 36 microbial species, 5 SCFAs, and 64 metabolites clustered into 7 communities within the network (Figure 6A). Butyrate, isovalerate, and Clostridium sensu stricto 1 sp. clustered into the yellow-green community with positive associations between their temporal changes between infancy and mid-childhood. Streptococcus sp. and histamine clustered together in the purple community with negative associations between them (Figure 6A), Among children who developed peanut allergy, 31 microbial species, 2 SCFAs, and 74 metabolites grouped into 10 communities. Butyrate, Clostridium sensu stricto 1 sp., and Streptococcus sp. clustered in the light-blue community (Figure 6B). Interestingly, although butyrate and Clostridium sensu stricto 1 sp. clustered together in both networks, the association between their temporal changes was positive in the peanut allergy-free network but negative in the peanut allergy development network. Another contrast observed was the differential relationships of change in Streptococcus sp. abundance within each of the networks, which clustered and had a negative association with histamine in the peanut allergy-free network, but exhibited no such association in the peanut allergy development network.

Figure 6. Integrated networks of gut microbiota and metabolites temporally associated with the development of peanut allergy.

Microbial-metabolite communities determined by xMWAS are indicated by node color. Microbial species and metabolites with temporal changes associated with the development of peanut allergy are labeled. Edges within a path length of two of labeled nodes in the same community are colored to indicate positive or negative relationships. Thicker edges highlight direct connectivity between labeled nodes. (A) Network for children who did not develop peanut allergy. All available microbiota and SCFA data from 87 NPA participants and global metabolite levels associated with peanut allergy development from 10 NPA participants were used to build the network. (B) Network for children who developed peanut allergy by mid-childhood. All available microbiota and SCFA data from 35 PA participants and global metabolite levels associated with peanut allergy development from 10 PA participants were used to build the network.

Discussion

In this longitudinal study of a unique, well-phenotyped, multi-center early-life cohort that followed infants at risk for peanut allergy through mid-childhood, we uncovered several new findings about the gut microbiome, metabolome, and their temporal dynamics associated with the development of peanut allergy. Such longitudinal, multi-modal, and integrated insight can be challenging to attain, as many prior studies have been cross-sectional, single center, based on looser definitions of allergy, and/or focused on the microbiome without integration of metabolite measures. The strengths of this study include its longitudinal study of a robustly phenotyped multi-center cohort with dedicated parallel profiling and integrated analysis of gut microbiome and metabolome measures.

We found that infants bound to develop peanut allergy had lower gut microbiome diversity, and that temporal change in species diversity was significantly associated with the development of peanut allergy (Figure 2). Microbiome diversity involves multiple dimensions, including species richness, evenness, and community composition. Large numbers of evenly distributed species characterize healthy, stable communities.^{29, 30} The significantly lower species evenness of gut microbiota that we observed in peanut allergy-bound infants suggests that they had less stable gut communities during this early window of rapid immune development. Although richness is often discussed, the other dimension of diversity captured by species evenness has been increasingly observed to affect ecosystem functioning more than previously appreciated.^{31, 32} Murine models have demonstrated that gut diversity in early life plays a pivotal role in reducing allergy risk.¹¹

Regarding species composition, we found that the relative abundances of specific gut species had significantly different temporal trajectories in children who did and did not develop peanut allergy (Figure 3). The longitudinal design of our study enabled identification of temporal patterns not detectable by prior cross-sectional studies. Changing abundances of Clostridium sensu stricto 1 sp. and Bifidobacterium sp. were each associated with the development of peanut allergy. Clostridium sensu stricto produces butyrate, a SCFA associated with reduced allergy-related inflammation and lower food allergy,^{14, 33–37} had low relative abundance in infants who later developed peanut allergy (Figure 3A), raising the possibility that its low abundance in early life contributes to a milieu that heightens risk for food allergy development. However, butyrate is produced by over 170 bacterial taxa,³⁸ so the relative abundance of Clostridium sensu stricto alone is unlikely to drive overall butyrate levels. Although over time Clostridium sensu stricto abundance increased in peanut allergy-bound children and decreased in children who remained peanut allergy-free (Figure 3C), there were no significant differences in its relative abundance once peanut allergy was established at mid-childhood.

We also observed that between infancy and mid-childhood, children who developed peanut allergy had decreasing relative abundance of Bifidobacterium sp., a member of a genus commonly used as probiotics to reduce allergy risk³⁹, while children who remained peanut-allergy-free had stable abundances of this species over time (Figure 3C). Murine models have shown that Bifidobacterium sp. can induce apoptosis of mast cells, reducing food allergy symptoms⁴⁰. Decreasing abundance of this ostensibly protective species over time was observed in children who developed peanut allergy (Figure 3C), such that by mid-childhood, the relative abundance of Bifidobacterium sp. was significantly lower among those with peanut allergy (Figure 3B). In contrast, Streptococcus sp. was higher at infancy in children who would go on to develop peanut allergy (Figure 3A), but it had no significant temporal change in abundance associated with peanut allergy development (Figure 3C). Streptococcus in the infant gut microbiome has been previously associated with another type of food allergy, IgE-mediated cow’s milk allergy.⁴¹

Gut microbiota exert influence on host immune function via metabolites they produce,^{14, 33, 42} and here again longitudinal analyses revealed temporal dynamics in SCFA levels associated with peanut allergy development undetected by cross-sectional analyses alone (Figure 4). Decreases in both butyrate and isovalerate between infancy and mid-childhood were associated with the development of peanut allergy (Figure 4C). Murine and in vitro models have shown that butyrate increases gut barrier integrity and tolerogenic cytokines, and reduces allergic responses by inhibiting Th2 cytokines and upregulating T regulatory mechanisms.^{34, 36} Infants who developed peanut allergy had decreasing butyrate levels over time, while those who remained peanut allergy-free had increasing levels (Figure 4C). Infants who developed peanut allergy also had decreasing isovalerate over time compared to stable levels in their peanut allergy-free counterparts (Figure 4C). Isovalerate is a branched SCFA that may also be protective against allergy given its previous associations with low vs. high sensitization in asthma⁴³ and farm life, where children have less atopy.⁴⁴ Its observed decrease over time in the children who developed peanut allergy is consistent with these prior works.

Interestingly, our pathway analysis of the global metabolome showed that metabolites with temporal changes associated with peanut allergy development clustered into a single pathway for histidine metabolism (Figure 5). Histidine skews cell differentiation toward atopy⁴⁵ and is the precursor to histamine, a bioactive amine released from mast cells during allergic reactions and a hallmark effector of allergic reactions⁴⁶. Histamine is also synthesized and secreted by many bacteria in the gut^47–49. Our results suggest that infants who go on to develop peanut allergy harbor gut microbial communities that produce histidine-pathway metabolites.

Network analysis facilitated integration of the multi-faceted findings from this study. Overall, the microbiota and metabolites associated with peanut allergy development demonstrated higher connectivity in the network for children who did not develop peanut allergy (Figure 6A) compared to those who did (Figure 6B). Previous work suggests that communities with greater numbers of highly connected nodes are less susceptible to perturbation by host or environmental factors.^50–52 Among children who remained peanut allergy-free (Figure 6A), changes in the relative abundance of Clostridium sensu stricto 1 sp. (a species associated with protection from allergy^{14, 33–37}) positively correlated with changes in both butyrate and isovalerate (SCFAs associated with protection from allergy^{34, 36, 43, 44}). In contrast, changes in the relative abundance of the species was negatively correlated with butyrate in children who developed peanut allergy (Figure 6B). This suggests that the same species may have different functional associations depending on phenotype. The relationships of Streptococcus sp. with metabolites also differed between children who did and did not develop peanut allergy. Its change in relative abundance over time negatively correlated with histamine in children who remained free of peanut allergy (Figure 6A), while such a correlation was absent in those who developed peanut allergy (Figure 6B).

We recognize limitations to this study. Our investigation focused on children with risk factors for peanut allergy development, and although we controlled for these factors in our models, it is possible that distinct or additional findings would be identified in a lower risk population. Environmental factors not measured in this study may have also influenced microbiome features and peanut allergy development. Although this study represents a major step forward with its dual examination of the gut microbiome and metabolome at infancy and mid-childhood, future study in larger cohorts with higher frequency sampling during the early years of life and global metabolomic profiling of greater numbers of samples could enable higher power for further detection of microbe-metabolite relationships associated with the development of peanut allergy. Additionally, longitudinal investigation of gut microbiome dynamics in infants selected from the general population could test the generalizability of this study’s findings to children without peanut allergy risk factors.

In summary, the findings from this integrative longitudinal study provide new insights into the dynamics of the gut microbiome, metabolome, and their relationship as children develop peanut allergy. Our findings from this unique and rigorously phenotyped multi-center early-life cohort reach beyond prior single-center, cross-sectional, and/or single data type studies of peanut allergy development. We found that species diversity at infancy and temporal changes in the relative abundance of specific microbiota significantly differ in children who develop peanut allergy. Our results lend support to butyrate, isovalerate, and histamine-influencing microbiota as mechanisms by which gut microbiota shape food allergy risk. The results from this integrated longitudinal study provide valuable insight that can inform our thinking on the mechanisms underlying and strategies for potentially preventing peanut allergy.

Key messages.

• Lower infant gut microbiome diversity was associated with peanut allergy development.

• Temporal changes in the relative abundance of specific microbiota and gut metabolite levels significantly differed in children who developed peanut allergy.

• Metabolites associated with peanut allergy development clustered within the histidine metabolism pathway.

Abbreviations

ASV

amplicon sequence variant

CoFAR2

Consortium for Food Allergy Research Observational Study

FDR

false discovery rate

HESI

heated electrospray ionization mode

LCMS

liquid chromatography tandem mass spectrometry

LME

linear mixed effect

LRT

likelihood ratio test

NPA

no peanut allergy

PA

peanut allergy

SCFA

short chain fatty acid

SD

standard deviation

SPT

skin prick test

Data availability

Data have been deposited in the NCBI Sequence Read Archive (BioProject ID: PRJNA868282) and the Metabolomics Workbench (Project ID PR001512).