Early detection of ureteropelvic junction obstruction in neonates with prenatal diagnosis of renal pelvis dilatation using 1H NMR urinary metabolomics

Abstract

Renal pelvis dilatation (RPD) is diagnosed in utero on prenatal ultrasonography (US) and can resolve spontaneously. However, isolated RPD can also reflect ureteropelvic junction obstruction (UPJO), which requires surgical treatment to prevent progressive renal deterioration. The diagnosis of UPJO can only be confirmed after birth with repeat US and renal isotope studies. ¹H Nuclear Magnetic Resonance spectroscopy (NMR) was performed on urine of newborns with prenatally diagnosed unilateral RPD and healthy controls to identify specific urinary biomarkers for UPJO. The original combination of EigenMS normalization and sparse partial-least-squares discriminant analysis improved selectivity and sensitivity. In total, 140 urine samples from newborns were processed and 100 metabolites were identified. Correlation network identified discriminant metabolites in lower concentrations in UPJO patients. Two main metabolic pathways appeared to be impaired in patients with UPJO i.e. amino acid and betaine metabolism. In this prospective study, metabolic profiling of urine samples by NMR clearly distinguishes patients who required surgery for UPJO from patients with transient dilatations and controls. This study will pave the way for the use of metabolomics for the diagnosis of prenatal hydronephrosis in clinical routine.

Figure 1

PCA and SPCA analysis of metabolites in urine after ¹H NMR analysis from TD and UPJO patients. (A) PCA scores with 87% of the variance explained on the first component and 2.1% on the second component. (B) SPCA scores with 86% of the variance explained on the first component and 2.2% on the second component. Fifty variables are selected on each component. SPCA allows the separation between groups. The different groups of samples are represented by colors: control (blue), TD (green), UPJO (red). The size of the dots varies according to the age of the patient in days (10 to 115). (C) Loadings plot from PCA with all variables and (D) Loadings plot from SPCA with 50 variables on each principal component.

Figure 2

SPLS-DA analysis of UPJO vs TD, UPJO vs controls and TD vs controls. (A) SPLS-DA scores plot PC1 (15%) vs PC2 (4%) with a good separation between UPJO (red) vs TD (green) ; (C) SPLS-DA scores plot PC1 (14%) vs PC2 (5%) with a good separation between UPJO (red) vs controls (blue); (E) SPLS-DA scores plot PC1 (25%) vs PC2 (6%) with a light separation between TD (green vs controls blue). For each model classification, a receiver operating characteristic (ROC) curve is shown (B,D,F).

Figure 3

(A) Average of NMR spectra from UPJO (red) and control (blue) and bucket’s annotation from SPLS-DA. (B)–(C) Correlation network of the variables from NMR. (B) UPJO versus TD (C) UPJO versus controls. The color indicate link between variables—red for anti-correlation and green for correlation. Thickness of lines demonstrates the intensity of the correlation. Correlation network was performed using qgraph R package.