Preterm neonatal urinary renal developmental and acute kidney injury metabolomic profiling: an exploratory study

Abstract

Background: Acute kidney injury (AKI) staging has been developed in the adult and pediatric populations, but these do not yet exist for the neonatal population. Metabolomics was utilized to uncover biomarkers of normal and AKI-associated renal function in preterm infants. The study comprised 20 preterm infants with an AKI diagnosis who were matched by gestational age and gender to 20 infants without an AKI diagnosis.

Methods: Urine samples from pre-term newborn infants collected on day 2 of life were analyzed using broad-spectrum nuclear magnetic resonance (NMR) metabolomics. Multivariate analysis methods were used to identify metabolite profiles that differentiated AKI and no AKI, and to identify a metabolomics profile correlating with gestational age in infants with and without AKI.

Results: There was a clear distinction between the AKI and no-AKI profiles. Two previously identified biomarkers of AKI, hippurate and homovanillate, differentiated AKI from no-AKI profiles. Pathway analysis revealed similarities to cholinergic neurons, prenatal nicotine exposure on pancreatic β cells, and amitraz-induced inhibition of insulin secretion. Additionally, a pH difference was noted. Both pH and the metabolites were found to be associated with AKI; however, only the metabotype was a significant predictor of AKI. Pathways for the no-AKI group that correlated uniquely with gestational age included aminoacyl-t-RNA biosynthesis, whereas pathways in the AKI group yielded potential metabolite changes in pyruvate metabolism.

Conclusions: Metabolomics was able to differentiate the urinary profiles of neonates with and without an AKI diagnosis and metabolic developmental profiles correlated with gestational age. Further studies in larger cohorts are needed to validate these results.

Keywords: Acute kidney injury; Metabolomics; Multivariate analysis; NMR spectroscopy; Neonatal; Regression analysis; Renal development.

Figure 1

The unsupervised PCA (a) and supervised multivariate analysis OPLS-DA (b for differentiating neonates with an AKI diagnosis from those without an AKI diagnosis. The supervised analysis had an R2Xcum=0.39, R2Ycum=0.54, and a Q2cum=−0.189. The classification was 80% correct for the cases, 90% correct for the controls, and 85% correct for all samples with a Fisher probability score of 8.3E-6. The OPLS-DA model had an area under the curve of 0.93 (c). (d) The curated pathways important for differentiating the AKI diagnosis.

Figure 2

The urinary pH was observed to be different in the no-AKI and AKI QC pools by the imidazole chemical shift (a) in the NMR spectra. Box plot of the calculated pH of no-AKI and AKI urinary profiles (b). The mean pH of the no-AKI was 7.29, while the AKI urinary profiles were 7.53, p = 0.007.

Figure 3

ROC curves for the three logistic regression models of the urinary profiles to predict AKI. Model (a) included pH as the only predictor of AKI had an area of the curve (AUC) of 0.68. Model (b) included the metabotype only had an AUC of 0.93. Model (c) which included both the calculated pH and the metabotype had an AUC of 0.93.

Figure 4

The predicted vs observed OPLS models of gestational age of no AKI profiles (a) and AKI profiles (b). Each point represents one urinary profile, and the goodness of fit (R2) is noted in the upper left corner. The metabolites that model the gestational age prediction are listed in tables 4 and 5.

Figure 5

Curated pathways important to the correlation with gestational age that are unique to neonates without an AKI diagnosis (a), with an AKI diagnosis (b), and common to all neonates regardless of AKI diagnosis (c).