Plasma metabolomics identifies differing endotypes of recurrent wheezing in preschool children differentiated by symptoms and social disadvantage

Preschool children with recurrent wheezing are a heterogeneous population with many underlying biological pathways that contribute to clinical presentations. Although the morbidity of recurrent wheezing in preschool children is significant, biological studies in this population remain quite limited. To address this gap, this study performed untargeted plasma metabolomic analyses in 68 preschool children with recurrent wheezing to identify metabolomic endotypes of wheezing. K-means cluster analysis was performed on metabolomic dataset including a total of 1382 named and unnamed metabolites. We identified three metabolomic clusters which differed in symptom severity, exacerbation occurrence, and variables associated with social disadvantage. Metabolites that distinguished the clusters included those involved in fatty acid metabolism, fatty acids (long chain monounsaturated fatty acids, long chain polyunsaturated fatty acids, and long chain saturated fatty acids), lysophospholipids, phosphatidylcholines, and phosphatidylethanolamines. Pathway analyses identified pathways of interest in each cluster, including steroid metabolism, histidine metabolism, sphingomyelins, and sphingosines, among others. This study highlights the biologic complexity of recurrent wheezing in preschool children and offers novel metabolites and pathways that may be amenable to future study and intervention.


Features of cluster 2
Nineteen children were grouped into Cluster 2, termed "less advantaged, most symptomatic wheezing." Children in this cluster were 47.4% male and slightly younger, with a median age of 29 months and a median wheezing onset at 12 months of age.Sixty three percent of the children in this cluster reported a race other than white and 68.4% of the children were currently symptomatic.Children in this cluster, like those in Cluster 1, overall were socially disadvantaged, with 52.7% of children living in areas classified as having either "low" or "very low" childhood opportunity.At 12 months after enrollment, 78.9% of children in this cluster had an emergency department visit for wheezing and 89.5% received a prednisone burst for a wheezing exacerbation.www.nature.com/scientificreports/

Features of cluster 3
Fifteen children were grouped into Cluster 3, termed "more advantaged, least symptomatic wheezing." Children in this cluster were 46.7% male, with a median age of 31 months and a median wheezing onset at 12 months of age.The majority of children in this cluster reported white race and this cluster also had the highest prevalence of household Bachelor's degree attainment.Compared to Clusters 1 and 2, only 33.3% of the children in this cluster were currently symptomatic.Children in this cluster had least social disadvantage, with only 13.4% of children living in areas classified as having either "low" or "very low" childhood opportunity.Instead, 60% of children in this cluster lived in areas classified as having either "high" or "very high" childhood opportunity.At 12 months after enrollment, 46.7% of children in this cluster had an emergency department visit for wheezing and 60% received a prednisone burst for a wheezing exacerbation.

Differences in individual metabolites between the clusters
Partial least squares discriminant analysis (PLS-DA), which is a "supervised" version of principal components analysis, was performed on all 1382 plasma metabolites (named and unnamed) to visualize the variability and to assess the discriminatory ability of metabolite profiles for the cluster groupings 49 .With the PLS-DA, component 1 and component 2 explained 5.9% and 4% of the variance, respectively (Fig. 1B).Children in Cluster 1 and Cluster 2 also had more within-group variability than children in Cluster 3, as indicated by the size of the ellipse over the points (Fig. 1C).
To determine the specific plasma metabolites that differed between clusters, analysis of variance was performed using a strict Bonferroni-corrected alpha level of 0.05 to control type 1 error.At a significance threshold of 3.62 × 10 5 , 79 significant metabolites were identified.Thirteen metabolites were unnamed, and 66 metabolites were named.Named metabolites were predominantly lipids and are shown in the heatmap in Fig. 2. Metabolites (who were more advantaged and the least symptomatic).In contrast, children in Cluster 2 had higher concentrations of lysophospholipids and other lipids including phosphatidylcholines and phosphatidylethanolamines (Fig. 3).Boxplots of selected lipid metabolites are shown in Fig. 3.

Exploratory analyses of aeroallergen sensitization and inhaled corticosteroid use
Although aeroallergen sensitization and inhaled corticosteroid use were not different between groups, we performed exploratory analyses with these variables in the entire sample, irrespective of cluster assignment.For this analysis, individual known metabolites were compared with area under the curve (AUC) and log2 fold change.

Discussion
This study performed untargeted plasma metabolomic profiling in a well-characterized sample of preschool children with recurrent wheezing to identify biological pathways for future investigation.With unsupervised k-means clustering of the plasma metabolite data, we identified 3 clusters of preschool children with recurrent wheezing which differed in measures of social disadvantage and respiratory symptoms, but not in features of Type 2 inflammation.Children in Cluster 2 who were less advantaged, the most symptomatic and the most likely to have a wheezing exacerbation by 12 months, had several altered lipids, including lower fatty acids and higher lysophospholipids, phosphatidylcholines and phosphatidylethanolamines. LIMMA also identified differentially expressed steroid metabolites and down-regulated metabolites associated with histidine metabolism, sphingomyelins, and sphingosines in children in Cluster 2. Together, these analyses suggest that that several biological pathways are perturbed in preschool children with recurrent wheezing, which warrant further study.Although metabolomics studies have been performed in school-age children with asthma [51][52][53][54][55] , very few studies have included preschool children with recurrent wheezing.One study found that preschool children with recurrent and persistent wheezing, compared to children with transient wheezing, had differing urinary metabolomic profiles 56 .Similar to children in our Cluster 2 (who were the most symptomatic), that prior study also found differential expression of phospholipids and fatty acid metabolites in children with persistent versus transient wheezing, suggestive of lipid dysregulation 56 .A separate study of preschool children with recurrent wheezing found that children with Type 2 inflammation (allergic sensitization), compared to children without Type 2 inflammation (no allergic sensitization), had differential expression of several amino acid metabolism pathways in plasma, with lower pyruvate and fumarate and increased glutamine, valine and isobutyric acid 57 .
Another study also found pyrimidine metabolism, glycerophospholipid metabolism, and arginine biosynthesis as distinguishing pathways between children who wheezed persistently after respiratory syncytial virus infection compared to those children who did not 58 .However, those previous studies included a small number of participants and utilized different biological samples for the metabolomic analyses.Our observation of altered fatty acids and lipids in the most symptomatic children in Cluster 2 also has biologic plausibility.Our finding of lower long chain polyunsaturated fatty acids in Cluster 2 supports other literature demonstrating that polyunsaturated fatty acids influence risk of allergic disease.For example, Lee-Sarwar et al. noted that plasma polyunsaturated fatty acid abundances were inversely associated with asthma and recurrent wheezing in children at three years of age 59 .Others have shown that after maternal supplementation with n-3 long chain polyunsaturated fatty acids during pregnancy, the child's metabolome at age 6 months displays several differences in lipid and amino acid pathways, with a decrease in metabolites in the n-6 long chain polyunsaturated fatty acid and tryptophan pathways and an increase in tyrosine pathways 60 .Children whose mothers received n-3 long chain polyunsaturated fatty acids also had a reduced risk of asthma by age 5 60 .Other lipid pathway alterations (i.e., glycerophospholipids, sphingolipids, and sphingolipid metabolism) have also been reported in asthma and are associated with lung function and airway hyperresponsiveness, while arachidonic acid and linoleic acid metabolism are associated with asthma control 61 .
Our observation of altered metabolomic profiles between less advantaged and more advantaged children is also potentially important.Despite the high prevalence of recurrent wheezing in preschool children, many of these children are from socially disadvantaged backgrounds [13][14][15] and this social disadvantage has been shown to influence wheezing outcomes 62 .For example, in the present study, the children in Clusters 1 and 2 with less advantage also likely have different access to food and different exposures, which can influence the metabolome.The comprehensive social impacts on the metabolome (referred to as the "exposome 63 ") have not been well studied but should be considered in future studies of recurrent wheezing in preschool children.
This study does have limitations.First, the statistical and clinical separation of the groups was modest, since thedifferences between the severity of the preschool wheeze are not that prominent between Cluster 1 and Cluster 2. Second, the young age of the children prohibited fasting, so the observed differences in metabolites between the cluster groups may not be solely due to altered catabolism of macromolecules.We cannot exclude other confounding since the clusters differed in several demographic variables including sex and race.The sample size was admittedly small, and this may have prevented detection of other metabolites of importance.Interestingly, some of the metabolites that provided the greatest discrimination between the cluster groups were also unmatched.Therefore, pathway enrichment analyses were not performed on all metabolites because many of these could not be mapped to the existing metabolomics databases.Future studies should place emphasis on less well characterized molecules for potential pathway discovery.We were limited by the cross-sectional analyses and cannot comment on the temporal stability of the metabolome in the cluster groups.Therefore, the specific differentiating metabolites we identified in this study require independent validation.Nonetheless, this work highlights the biologic complexity of recurrent wheezing in preschool children and offers novel metabolites and pathways that may be amenable to future study and intervention.

Methods
Preschool children 12-59 months of age with recurrent wheezing, defined as a lifetime history of two or more episodes of wheezing, each lasting at least 24 h and requiring repeated treatment with albuterol sulfate, were approached for the study at an outpatient respiratory clinic at Children's Healthcare of Atlanta, in Atlanta, Georgia.Children were included in the study if they had at least one wheezing episode treated with systemic corticosteroids in the previous 12 months.Children who received systemic corticosteroids within 4 weeks prior to enrollment were rescheduled.Exclusion criteria included premature birth before 35 weeks gestation, immune deficiency, cystic fibrosis, pulmonary aspiration, congenital airway anomalies, and failure to thrive.The Emory University/Children's Healthcare of Atlanta Institutional Review Board approved the study protocol and informed written consent was obtained for the study from the child's parent prior to study participation.All study procedures were performed in accordance with the relevant guidelines and regulations in the Declaration of Helsinki.

Study design and characterization procedures
Preschool children attended a research-only outpatient visit that was rescheduled if the child had an acute illness or a wheezing exacerbation treated with systemic corticosteroids in the preceding two weeks.At the research visit, caregivers completed demographic questionnaires and medical history questionnaires.Respiratory symptom severity was assessed over the previous week with five questions on the severity of cough, wheezing, trouble breathing, activity interference and sleep interference, each of which was scored from zero (none) to five (very severe) and summed, with higher scores reflecting greater respiratory symptoms.Preschool children also submitted blood samples for plasma metabolomics, as well as total serum IgE and blood eosinophil counts, which were performed at a hospital laboratory (Children's Healthcare of Atlanta, Atlanta, Georgia).At the completion of the visit, the children were followed for 12 months.Outcomes at 12 months included the occurrence of any wheezing episode resulting in an emergency department visit or any wheezing exacerbation requiring treatment with prednisone or equivalent systemic corticosteroids.

Residential geocoding for childhood opportunity
Residential geocoding was performed as described previously 64 by mapping participant residential addresses to census tracts using the R package tidygeocoder (R version 4.0.2) 65 .U.S. Census 2020 Geographic Identifiers were mapped to the 2010 GEOIDs for each census tract in Georgia using the R package tigris 66 .Socioeconomic data for each census tract was obtained using the R package tidycensus 67 .The COI 2.0 for 2015 48 , which provides a measure of neighborhood resources and conditions that affect child development, was downloaded and joined to patient data using the census tract GEOID 68 .The COI 2.0 consists of 29 indicators grouped into three domains: education, health and environment, and social and economic resources and opportunities.The COI 2.0 rankings at the level of the state of Georgia were used in the analysis.

Plasma metabolomics
Blood for metabolomics was collected by venipuncture into an ethylenediaminetetraacetic acid vacutainer tube and centrifuged at 400XG to separate cells from platelet-rich plasma.After initial separation, plasma was centrifuged at 3000XG for 10 min at 4 °C to remove any remaining red blood cells or platelets prior to being frozen at − 80 °C.Plasma metabolomics data acquisition was performed at Metabolon (Morrisville, NC).Sample proteins were precipitated with methanol under vigorous shaking for 2 min (GenoGrinder 2000, Glen Mills, Clifton, NJ) followed by centrifugation.Samples were placed briefly on an evaporator (Zymark TurboVap, Marshall Scientific, Hampton, NH) to remove organic solvent.UPLC-Tandem MS was performed with a liquid chromatograph (Acquity, Waters, Milford, MA) and a high resolution/accurate mass spectrometer interfaced with a heated electrospray ionization source (Q-Exactive, Thermo Scientific, Waltham, MA) and Orbitrap mass analyzer operated at 35,000 mass resolution 69 .The first aliquot was analyzed using acidic positive ion conditions, chromatographically optimized for more hydrophilic compounds (positive early platform).In this method, the extract was gradient eluted from a C18 column (Waters UPLC BEH C18-2.1 × 100 mm, 1.7 µm) using water and methanol, containing 0.05% perfluoropentanoic acid and 0.1% formic acid.The second aliquot was also analyzed using acidic positive ion conditions but was chromatographically optimized for more hydrophobic compounds (positive late platform).In this method, the extract was gradient eluted from the same aforementioned C18 column using methanol, acetonitrile, water, 0.05% perfluoropentanoic acid and 0.01% formic acid and was operated at an overall higher organic content.The third aliquot was analyzed using basic negative ion optimized conditions using a separate dedicated C18 column (negative platform).The basic extracts were gradient eluted from the column using methanol and water, however with 6.5 mM Ammonium Bicarbonate at pH 8.The fourth aliquot was analyzed via negative ionization following elution from a Hydrophilic Interaction Liquid Chromatography column (Waters UPLC BEH Amide 2.1 × 150 mm, 1.7 µm) using a gradient consisting of water and acetonitrile with 10 mM Ammonium Formate, pH 10.8 (polar platform).The MS analysis alternated between MS and datadependent MS n scans using dynamic exclusion.The scan range varied slightly between methods but covered 70-1000 m/z.Experimental samples were randomized across the platform run with quality control samples spaced evenly among the injections.Peaks were quantified using area-under-the-curve.

Data analyses
Metabolomics data analyses were performed with R software (version 4.2.3) 70.Metabolomics data were logarithmically (log2) transformed and then underwent UMAP for dimension reduction 47 , yielding two UMAP coordinates for each participant.K-means clustering was performed based on UMAP coordinates and an elbow plot was used to determine the optimal k number of clusters.The log2-transformed metabolites were then quantile normalized and PLS-DA was performed using MetaboAnalyst 5.0 49 .Significant metabolites that differed www.nature.com/scientificreports/ between clusters were assessed with general linear models using a Bonferroni-corrected alpha level of 0.05 to control type 1 error.Differential expression of metabolites in each cluster was assessed with the software package LIMMA, which employs a moderated t-statistic, known as parametric Empirical Bayes methodology, for differential expression 50 .Pathway analysis was performed with MetaboAnalyst 5.0 49 on log-transformed (base 10) and quantile normalized metabolites without data scaling using global test enrichment methods, and relativebetweenness centrality.Pathway analysis integrated pathway enrichment analysis and pathway topology analysis and was restricted to plasma metabolites that could be mapped to the Human Metabolome Database 71 .Other data analyses, including cluster group comparisons of clinical features, were performed with chi-square tests or general liner models with Fisher's Least Significant Difference post hoc tests, with a significance threshold of 0.05.

Figure 1 .
Figure 1.Cluster identification.(A) K-means clustering of 1382 plasma metabolites after Uniform Manifold Approximation and Projection (UMAP) for dimension reduction identifies three metabolomic clusters.Individual children are shown as circles (children without aeroallergen sensitization) or as + symbols (children with aeroallergen sensitization).(B) Silhouette scores for varying cluster solutions.(C) Partial least squares discriminant analysis (PLS-DA) results from all 1382 plasma metabolites, demonstrating the discriminatory ability of metabolite profiles for the cluster groupings.

Figure 2 .
Figure 2. Heatmap of significant metabolites that differ between cluster groups.Normalized concentrations of significant metabolites that differed between the three cluster groups at a Bonferroni-adjusted significance threshold of 0.05 are shown for each cluster group.Higher normalized concentrations are shown in orange and lower normalized concentrations are shown in blue.Metabolites designated with an "X" are unnamed.

Figure 4 .
Figure 4. Differential expression of metabolites in each cluster identified by Linear Models for Microarray Data (LIMMA).(A) Volcanco plots are shown for each cluster, with upregulated and downregulated metabolites shown in red and blue, respectively.(B) Overlap of differentially expressed metabolites between cluster groupings.

Figure 5 .
Figure 5. Metabolic pathway analyses.The results of metabolic pathway analyses for (A) cluster 1, (B) cluster 2, and (C) cluster 3. Pathway impact reflects the relative importance of the identified metabolites in the pathway.Larger circles reflect more important pathway impact.Significant pathways at a false discovery rate of 0.05 are shown in red.Non-significant pathways are shown in orange, yellow and white. https://doi.org/10.1038/s41598-024-66878-1

Table 2 .
Social disadvantage, measured by the residential Childhood Opportunity Index (COI) version 2.0.Data shown are the overall COI score, the three COI domain scores, and significant indicators (p < 0.05) within each domain.Data are presented as the median (25 th , 75 th percentile). 1 Higher COI overall scores and higher COI domain scores reflect better childhood opportunity.*p < 0.05 vs. Cluster 1; #p < 0.05 vs. Cluster 2.

Less advantaged, moderately symptomatic Less advantaged, most symptomatic More advantaged, least symptomatic
Vol:.(1234567890) Scientific Reports | (2024) 14:15813 | https://doi.org/10.1038/s41598-024-66878-1www.nature.com/scientificreports/MetabolicpathwayanalysesMetabolic pathway analyses were then performed for each cluster to understand how the differentially expressed individual metabolites relate to known metabolic pathways.With a false discovery rate of 0.05 as the threshold for significance, 11 pathways were significantly impacted in Cluster 1, which included children who were less