Benchmarking 2D NMR Metabolomics Identification

Twelve 2D NMR experiments (six TOCSY and six HSQC) were collected under different pH conditions using three synthetic mixtures and a plasma sample. The three synthetic mixtures were composed of 27, 21, and 24 common metabolites, respectively, with concentrations ranging from 40 to 60 mM. The plasma sample contained 35 identifiable metabolites (ranging in concentration from 0.1 to 10 mM) as determined by independent profiling of its 1D ¹H NMR spectra by several experienced individuals using Chenomx's NMR Suite software [12 (link)]. These results were further confirmed by spiking/doping authentic standards into the plasma sample and by GC-MS analysis. The plasma sample was prepared by filtering the sample through a 3 kDa filter (to remove proteins), then lyophilizing and finally dissolving the remaining solids in distilled water to its 1/5 original volume. Deuterium oxide (D₂O) was added to make a final concentration of 90% H₂O and 10% D₂O. All spectra were acquired at 25°C. Six spectra were collected on a Varian INOVA 800 MHz spectrometer equipped with a 5 mm triple axis gradient cryoprobe. The other six spectra were collected on a Varian INOVA 500 MHz spectrometer with a 5 mm triple-resonance z-gradient probe. The TOCSY experiments were performed using the wgtocsy pulse sequence, and the HSQC experiments were performed using the gChsqc pulse sequence, both provided by Varian's BioPack. For the TOCSY experiments, the spectral width was set to 11990 Hz and a mixing time of 50 milliseconds. Sixteen transients were collected for each of 128 t₁ intervals using an acquisition time of 0.085 seconds and a relaxation delay of 2.0 seconds. The total acqisition time for the TOCSY spectra was 2.5 hours For the ¹³C-HSQC experiments, the spectral widths of the proton and carbon dimensions were 11990 Hz and 28160 Hz respectively. Sixty four transients were acquired for each t₁ interval using an acquisition time of 0.085 seconds and a relaxation delay of 1.0 seconds. The spectra were collected with 2048*256 complex points for the ¹H and ¹³C dimensions respectively. The total spectral acquisition time for the HSQC spectra was 5 hours. Sample TOCSY and HSQC spectra are available [see Additional File 1].
The raw NMR spectra were first processed using NMRPipe [27 (link)] and the peaks were subsequently picked using Sparky's [28 ] automatic peak picking program. The resulting "raw" peak lists were copied and pasted to the processing view of MetaboMiner. Both peak processing and compound identification were performed using MetaboMiner's default parameter sets. The reference library used for the synthetic mixtures was the biofluid (common) library. For plasma data, the plasma (common) library was used. To assess the degradation in performance assuming no prior knowledge of the sample source (urine, plasma, cell extract or generic biofluid) the complete spectral reference library (223 compounds for TOCSY, 502 compounds for HSQC) was also used to identify compounds. To assess the performance of the web-servers that support 2D NMR mixture analysis – the HMDB [23 (link)], the MMCD [25 (link)], the BMRB [24 (link)], and the SpinAssign [29 ] of PRIMe – the same set of peak lists were submitted. For PRIMe, the default search parameters were used. For other web services, the search threshold for ¹H was set to 0.03 ppm and 0.10 ppm for ¹³C. The results are summarized in Tables 1 and 2.