How To Collect NMR Spectra for Optimal Performace with Bayesil

In order for Bayesil to perform optimally, it is best if the sample preparation and the NMR data acquisition are performed in a manner similar to that used to prepare and collect the reference NMR spectra used in Bayesil's spectral library. Using different spectral preparation and collection conditions will compromise the performance, particularly the quantification accuracy. Here we describe the recommended methods that Bayesil users should follow when preparing biological samples and collecting NMR spectra of those samples.

Sample Preparation

All samples should have been either stored frozen or stored at 4 °C prior to use. Cerebrospinal fluid should be spun down to remove particulate matter and its pH checked (or adjusted) to ensure that it is between pH 6.8-7.4. More attention must be paid for the sample preparation of serum or plasma because of the high concentration of proteins in these samples. Proteins in the biofluid sample of interest must be removed by passing it through a pre-rinsed (washed 7X) Amicon Ultra-0.5 3000 MWCO filter at 4 °C using a centrifuge (12,000 g force). Unwanted protein signals are particularly problematic with serum and plasma and ultrafiltration is currently the best method of removing these signals without altering the chemical composition of the biofluid. After the protein filtration step has been completed, the sample may be placed into a NMR tube. If Shegemi tubes are used, 285 µL of sample will be needed. For regular thin walled 5 mm NMR tubes, 570 µL of sample will be needed. With respect to pre-rinsing the Amicon filters, sterile double-distilled water or its equivalent should be used. Rinsing should be done at least seven times to ensure any residual glycerol is removed. Otherwise glycerol concentrations will be erroneously elevated in the biofluid sample.

It is recommended that users check the pH of each sample prior to spectrum collection to ensure the pH is near 7.0. Otherwise the global fitting routine will produce larger errors than reported. The physiologic pH of blood and CSF is between 7.3 and 7.4. Plasma and serum generally have large buffering capacities, thus pH adjustment is rarely problematic. Bayesil is designed to work well with samples having a pH range between 6.8 and 7.5. However, the pH may vary due to sample handling or disease state and adequate optimization of the pH prior to data collection cannot be stressed enough. Larger concentrations of sodium phosphate than 50 mM may be necessary if your sample's pH is not adequately controlled. Thus, checking your sample's pH beforehand can alleviate many potential problems.

The necessary volumes and buffer recipes required for sample preparation are shown below in Table 1 and Table 2 respectively.

Table 1: Sample components.

Sample SizeBiofluid VolumeD2OBufferFinal VolumeNMR tube
350 µl285 µl35 µl30 µl350 µlShegemi
700 µl570 µl70 µl60 µl700 µlRegular 5 mm

Table 2: 11.7x Buffer Recipe.

Ingredients11.7x Stock (100 ml)ConcentrationFinal Concentration
Na2HPO41 (anhydrous)7.850 g583 mM50 mM
DSS20.230 g11.7 mM1.0 mM

Table 3: Alternate 11.7x Buffer Recipe.

Ingredients11.7x Stock (100 ml)ConcentrationFinal Concentration
K2HPO41 (anhydrous)30.450 g1.75 M150 mM

Sodium Formate2


2-chloro pyrimidine-5-carboxylic acid2

0.0079 g


0.0926 g

1.167 mM


5.84 mM

100 uM


500 uM

DSS-d630.115 g5.833 mM0.5 mM
pH should be adjusted to ~7.0 with NaOH and/or HCl as needed. The standard samples and biological samples used in this study were prepared with a final sodium phosphate concentration of 20 mM. More recent metabolomic studies with different biological samples have yielded insufficient pH control at this concentration and upwards of 150 mM may be required for adequate pH adjustment of urine or CSF samples. Potassium phosphate buffers have better aqueous solubility at higher molarity and are a good substitution for sodium phosphate if concentrated stock solutions need to be prepared (see Table 3).
The inclusion of 100 uM formate provides a down field signal that greatly improves Bayesil's phasing result during automatic spectral processing. Poorly phased spectra give rise to quantification errors due to erroneous area calculations under each peak. Because formate can be found in some biological samples, care should be taken to accurately weigh out the ingredient as its concentration will need to be subtracted from the reported value. For accuracy, larger, more concentrated volumes should be prepared in volumetric flasks and diluted to the required value. Alternatively, we have found the non-biological compound 2-chloropyrimidine-5-carboxylate to be an acceptable alternative. It is stable at room temperature for more than 1 week and shifts at 8.76 ppm, which does not overlap with any biological metabolites we have been able to identify in either serum or CSF.
Various acronyms of DSS include: 4,4-Dimethyl-4-silapentane-1-sulfonic acid, 3-(Trimethylsilyl)propane-1-sulfonic acid or 2,2-dimethyl-2-silapentane-5-sulfonate. d6 refers to the perdeuterated form.

NMR Spectral Acquisition Parameters

All one-dimensional 1H-NMR spectra should be collected using the first increment of a standard two-dimensional presaturation NOESY experiment (see pulse sequence below).

Pulse sequence

With Agilent/Varian instruments, this experiment can be found in the BioPack experimental library and has the pulse sequence label “tnnoesy”. The comparable Bruker experiment is CPPWaterSupp (pulse program: pusenoesypr1d). All NMR spectra should be collected at 25 °C. A 12 ppm sweep width, a 4 s acquisition time, a 100 ms mixing time, a 10 ms recycle delay and a 990 ms saturation delay should be used. Typically, 128 transients should be acquired for samples collected at 500 MHz or 600 MHz, although this may vary depending on the quality of the probe (cold/cyroprobe versus room temperature probe). Eight steady-state scans should collected prior to data acquisition and the pre-saturation pulse power should be calibrated to provide a field strength no greater than 60 to 80 Hz. This minimizes the saturation of signals near the water resonance (4.78 ppm). In addition to the presaturation pulse calibration, both the transmitter offset and the saturation pulse should be positioned on the water resonance. No other solvent suppression pulses, including gradients, should be used (i.e. all gradient pulse widths and powers should be set to zero). Standard phase cycling should be used to suppress the water resonance and other spectral artifacts. Thus, it is imperative that the pulse sequence be set up such that the number of transients is a multiple of the number of pulses in each cycle. For the Varian/Agilent sequence, it is 16. All excitation pulses should be calibrated to correspond to a 90° flip angle (7.7 µs for the probe used in the above illustrated pulse sequence).


Mercier, P., Lewis, M.J., Chang, D., Baker, D. and Wishart, D.S. (2011) Towards automatic metabolomic profiling of high-resolution one-dimensional proton NMR spectra J. Biomol NMR 49:307–323