Neonatal Urine Metabolic Profiling and Development of Childhood Asthma

Abstract

Urine metabolomics case-control studies of childhood asthma have demonstrated a discriminative ability. Here, we investigated whether urine metabolic profiles from healthy neonates were associated with the development of asthma in childhood. Untargeted metabolomics by liquid chromatography-mass spectrometry was applied to urine samples collected at age 4 weeks in 171 and 161 healthy neonates born from mothers with asthma from the COPSAC2000 and COPSAC2010 cohorts, respectively, where persistent wheeze/asthma was prospectively diagnosed using a symptom-based algorithm. Univariate and multivariate analyses were applied to investigate differences in metabolic profiles between children who developed asthma and healthy children. Univariate analysis showed 63 and 87 metabolites (q-value < 0.15) in COPSAC2000 and COPSAC2010, respectively, which is promising for discriminating between asthmatic and healthy children. Of those, 14 metabolites were common among the two cohorts. Multivariate random forest and projection to latent structures discriminant analyses confirmed the discriminatory capacity of the metabolic profiles in both cohorts with estimated errors in prediction equal to 35% and AUCpred > 0.60. Database search enabled annotation of three discriminative features: a glucoronidated compound (steroid), 3-hydroxytetradecanedioic acid (fatty acid), and taurochenodeoxycholate-3-sulfate (bile acid). The urine metabolomics profiles from healthy neonates were associated with the development of childhood asthma, but further research is needed to understand underlying metabolic pathways.

Keywords: asthma; childhood; metabolomics; neonate; urine; wheezing illnesses.

Figure 1

ROC curve for the COPSAC2010 dataset. In the plot, the 100 ROC curves calculated during the projection to latent structures discriminant analysis based on variable influence on projection selection (VIP-based PLS-DA) stability selection procedure for the COPSAC2010 dataset are reported, represented by the black lines. Red, dashed lines indicate the 10th, 50th, and 90th percentiles.

Figure 2

Random forest analysis of the COPSAC2010 dataset. Error-out-of bag (errOOB) versus the number of trees of the 500 random forests built on the COPSAC2010 dataset. Black, dashed lines indicate the 10th, 50th, and 90th percentiles of the errOOB. Red lines indicate the errOOB for the non-asthma group, the green lines that of the asthma group, and the black lines the total errOOB for each random forest. The total errOOB laid in the range of 0.28 to 0.39 after 300 trees. With more than 300 trees, the performance did not change.