Experimental design of blending samples to increase precision of 1H NMR urine metabolomics

Urine is an equally attractive biofluid for metabolomics analysis as it is a challenging matrix analytically. Accurate urine metabolite concentration estimates by Nuclear Magnetic Resonance are hampered by pH and ionic strength differences between samples resulting in large peak shift variability. Here we show that calculating spectra of original samples from mixtures of samples using linear algebra reduces the shift problems and makes various error estimates possible. Since the use of 2D NMR to confirm metabolite annotations is effectively impossible to employ on every sample of large sample sets, stabilization of metabolite peak positions increases the confidence in identifying metabolites, avoiding the pitfall of oranges-to-apples comparisons.