Nasopharyngeal lipidomic endotypes of infants with bronchiolitis and risk of childhood asthma: a multicentre prospective study

Abstract

Background: Bronchiolitis is the leading cause of hospitalisation of US infants and an important risk factor for childhood asthma. Recent evidence suggests that bronchiolitis is clinically heterogeneous. We sought to derive bronchiolitis endotypes by integrating clinical, virus and lipidomics data and to examine their relationship with subsequent asthma risk.

Methods: This is a multicentre prospective cohort study of infants (age <12 months) hospitalised for bronchiolitis. We identified endotypes by applying clustering approaches to clinical, virus and nasopharyngeal airway lipidomic data measured at hospitalisation. We then determined their longitudinal association with the risk for developing asthma by age 6 years by fitting a mixed-effects logistic regression model. To account for multiple comparisons of the lipidomics data, we computed the false discovery rate (FDR). To understand the underlying biological mechanism of the endotypes, we also applied pathway analyses to the lipidomics data.

Results: Of 917 infants with bronchiolitis (median age, 3 months), we identified clinically and biologically meaningful lipidomic endotypes: (A) cinical^classiclipid^mixed (n=263), (B) clinical^severelipid^{sphingolipids-high} (n=281), (C) clinical^moderatelipid^{phospholipids-high} (n=212) and (D) clinical^atopiclipid^{sphingolipids-low} (n=161). Endotype A infants were characterised by 'classic' clinical presentation of bronchiolitis. Profile D infants were characterised by a higher proportion of parental asthma, IgE sensitisation and rhinovirus infection and low sphingolipids (eg, sphingomyelins, ceramides). Compared with endotype A, profile D infants had a significantly higher risk of asthma (22% vs 50%; unadjusted OR, 3.60; 95% CI 2.31 to 5.62; p<0.001). Additionally, endotype D had a significantly lower abundance of polyunsaturated fatty acids (eg, docosahexaenoic acid; FDR=0.01). The pathway analysis revealed that sphingolipid metabolism pathway was differentially expressed in endotype D (FDR=0.048).

Conclusions: In this multicentre prospective cohort study of infants with bronchiolitis, integrated clustering of clinical, virus and lipidomic data identified clinically and biologically distinct endotypes that have a significantly differential risk for developing asthma.Delete.

Keywords: asthma; paediatric asthma.

Figure 1.. Analytic workflow of lipidomic endotyping

The analytic cohort consists of 917 infants hospitalized for bronchiolitis in a multicenter prospective cohort study—MARC-35. At enrolment, the nasopharyngeal airway specimens were collected for lipidomic and transcriptomic profiling. 1. Clustering individual datasets: We first computed a distance matrix and identified mutually exclusive clusters for each of the clinical, virus, and lipidomic datasets by applying consensus clustering algorithms. 2. Clustering fused matrix: By integrating these derived clusters from three datasets, we generated a fused matrix and computed a Gower distance. By applying a consensus clustering algorithm to the fused matrix, we identified four mutually exclusive lipidomic endotypes. To choose an optimal number of profiles, we used a combination of consensus matrix, consensus cumulative distribution function, cluster consensus value, endotype size, and clinical and biological plausibility. 3. Examining clinical and biological characteristics of endotypes: To interpret the lipidomic endotypes clinically and biologically, we developed chord diagrams on major clinical and virus variables and a heatmap of 14 major lipid classes. 4. Determining clinical significance of endotypes: To examine clinical significance of the endotypes, we determined the longitudinal relationship of the lipidomic profiles with the risk for developing asthma (the primary outcome) and recurrent wheeze (the secondary outcome). We constructed logistic regression models for asthma development and Cox proportional hazards models for recurrent wheeze. 5. Investigating the biological significance of each endotype: We conducted three analyses: i) between-endotype examinations of fatty acids, ii) metabolic pathway analysis, and iii) integrated transcriptomic-lipidomic pathway analysis. 6. Examining the robustness of the findings: In the sensitivity analysis, we also examined the concordance between different numbers of lipidomic endotypes.

Figure 2.. Major clinical and virus characteristics according to the lipidomic endotypes among infants hospitalized for bronchiolitis

The ribbons connect from the individual lipidomic profiles to the major clinical and virus characteristics. The width of the ribbon represents the proportion of infants within the profile who have the corresponding clinical or virus characteristic, which was scaled to a total of 100%. For example, in panel C, the endotype A infants (blue) had a high proportion of males, a low proportion of previous breathing problems, parental history of asthma and eczema, and IgE sensitization, and a high proportion of RSV infection. In contrast, endotype D (red) infants had a high proportion of previous breathing problems, parental history of asthma and eczema, and rhinovirus infection. Abbreviations: IgE, immunoglobulin E; RSV, respiratory syncytial virus A. Comparison of endotype A (blue) with endotype B (green) B. Comparison of endotype A (blue) with endotype C (orange) C. Comparison of endotype A (blue) with endotype D (red)

Figure 3.. Lipid classes and fatty acids according to the lipidomic endotypes among infants hospitalized for bronchiolitis

Heatmaps show the mean values for the corresponding A) lipid classes and B) fatty acids in each of the four lipidomic endotypes. The areas of circles and colors represent the mean value of the corresponding value. Each variable is standardized by using auto-scaling. The between-endotype differences were examined by the Kruskal-Wallis test. * False discovery rate <0.05

Figure 4.. Association of the lipidomic endotypes among infants hospitalized for bronchiolitis with the risk for developing asthma and recurrent wheeze

* To examine the association of the lipidomic endotypes (endotype A as the reference) with the risk for developing asthma, a mixed-effects logistic regression model accounting for potential patient clustering within the hospitals was fit. ^† To examine the association between the endotypes (endotype A as the reference) and the rate of recurrent wheeze, a mixed-effects Cox proportional hazards model was constructed. Abbreviation: CI, confidence interval

Figure 5.. Kaplan-Meier curves for development of recurrent wheeze by age three years, according to the lipidomic endotypes, among infants hospitalized for bronchiolitis

Overall, the survival curves significantly differed between the endotypes (P_log-rank=0.01). Compared with endotype A infants (clinical^classiclipid^mixed), the rate of developing recurrent wheeze by age three years was not significantly different in endotype B or endotype C infants. In contrast, the rate was significantly higher in endotype D (clinical^atopiclipid^{sphingolipids-low}) infants (HR 1.65; 95% CI 1.19–2.27; P=0.003). The corresponding hazard ratio estimates are presented in Figure 4.

Figure 6.. Between-endotype differences (A vs. D) in lipidomic profiles and pathways among infants hospitalized for bronchiolitis

A) Lipid classes: The mean values for the corresponding lipid classes in each of lipidomic endotypes (A vs. D) are plotted. Each lipid class is standardized by using auto-scaling. The between-endotype differences were examined by Wilcoxon rank-sum test. B) Fatty acids: The mean values for the corresponding fatty acids in each of lipidomic endotypes (A vs. D) are plotted. Each fatty acid is standardized by using auto-scaling. The between-endotype differences were examined by Wilcoxon rank-sum test. C) Metabolic pathway analysis: In the metabolite pathway analysis, all detected pathways (based on the Kyoto Encyclopedia of Genes and Genomes [KEGG]) are shown. The color of each dot represents the pathway impact. The pathway impact is calculated as the sum of the importance measures of the matched metabolites normalized by the sum of the importance measures of all metabolites in each pathway [59]. D) Integrated transcriptomic-lipidomic pathway analysis: In the integrated transcriptomic-lipidomic pathway analysis, 20 pathways (based on KEGG) with the lowest FDRs are selected. The color of each dot represents the ratio of hit lipids and genes for the corresponding pathways. * False discovery rate (FDR) of <0.05 and ** FDR of ≤0.10 in panels A and B. Abbreviations: DHA, docosahexaenoic acid; EPA, eicosapentaenoic acid; GPI, glycosylphosphatidylinositol; KEGG, Kyoto Encyclopedia of Genes and Genome