Asthma is a heterogenous condition that remains poorly understood. Current guidelines do not sufficiently capture this heterogeneity, leading to suboptimal management and treatment strategies. A more comprehensive classification of asthma into biologically meaningful subgroups is needed.

We performed plasma metabolomic profiling of 1155 asthmatic children across four platforms covering a broad range of the metabolome; C8-positive, HILIC-positive, C18-negative and targeted Amide-Negative. We generated patient similarity networks for each platform that connected asthmatics via edges representing the similarity in their metabolomic profiles (controlling for age, sex and body mass index). We then fused the four networks using Similarity Network Fusion and performed spectral clustering on the resulting fused network.

We identified four clusters, or metabo-endotypes, and determined there was a significant difference across the metabo-endotypes in asthma relevant phenotypes including lung function (FEV1/FVC ratio, p-ANOVA=1.7x10-6); eosinophil count (p-ANOVA=0.04) and prevalent hay fever (p-ANOVA=0.01) a common asthma co-morbidity. We then recapitulated the four metabo-endotypes using plasma metabolomic profiles from an independent population of childhood asthmatics (n=955) and demonstrated these recapitulated metabo-endotypes shared the same clinical features as the discovery endotypes, with significant differences (p<0.05) in the same asthma phenotypes. We further observed that the metabo-endotype defined by the mildest asthma, was associated with higher levels of anti-inflammatory metabolites, while the opposite was seen in the metabo-endotype with the most severe asthmatics.

These findings demonstrate that clinically meaningful endotypes can be derived and validated using metabolomic data, and that interrogating the drivers of these metabo-endotypes can help understand their pathophysiology and generate therapeutic targets.