Acquisition and analysis of 1H NMR

¹H NMR spectra were acquired with a Bruker Avance III 400 MHz spectrometer operating at 400.13 MHz for ¹H at 310 K equipped with a 5 mm broad-band inverse configuration probe. Samples were analysed in random order and automation with a SampleCase 24 sample automation system. Samples were analysed using water suppressed 1D NMR spectrum using the NOESYPRESAT pulse sequence (160 transients) and a Carr-Purcell-Meiboom-Gill (CPMG) spin echo sequence with pre-saturation (160 transients). Irradiation of the solvent (water) resonance was applied during pre-saturation delay (2.0 s) for all spectra and for the water suppressed 1D NMR spectra also during the mixing time (0.1 s). The pulse sequence parameters including the 90° pulse (~ 12 µs), pulse frequency offset (~ 1880 Hz), receiver gain (90.5), and pulse powers were optimised on a representative sample and then set constant for the cohort analysed. The spectral width was 30 ppm for NOESY experiments and 20 ppm for CPMG experiments. The resulting Fourier induction decays were processed with an exponential line broadening of 0.3 Hz prior to Fourier transformation, which were collected with approximately 32 k real data points.

¹H NMR raw datasets were automatically phased, baseline corrected and referenced to the α-C₁H glucose doublet (5.23 ppm) using MATLAB 7.0 software (MathWorks, Natick, MA). To reduce analytical variation between samples, the residual water signal (4.67–4.98 ppm) was truncated from the data set. Probabilistic quotient normalisation was performed across the cohort²³. Assignment of endogenous metabolites was made to a high confidence using Chenomx^® (Chenomx Inc., Edmonton, AB, Canada) and by reference to published literature, online resources, and spiking experiments^24–26. Statistical Recoupling of Variables²⁷ was used to cluster pre-processed NMR spectrum data into features or peaks representing the relative metabolite abundance.