Metabolite analysis of fluids from the circulatory system provides a view of the metabolic state of an organism. Unlike urine analysis, which measures an organism’s waste products, serum or plasma analysis measures homeostatic levels of metabolites throughout the organism.

In theory, traditional chemometric approaches, such as spectral binning (also known as "spectral bucketing") and pattern recognition, can be applied to serum spectra. Practically, though, this can be problematic due to the substantial area contributed by protein resonances. The NMR spectrum of a serum sample includes both sharp, narrow peaks from small molecule metabolites and broad peaks from proteins and lipids. As a result, many methods of analyzing serum spectra rely on methods of removing the effects of protein resonances, after which standard analysis techniques may be used (see [16], [17] and Analysis Methods).

Several methods for removing removing protein resonances from serum spectra are available. NMR editing techniques use pulse sequences designed to exclude detection of specific types of signals. For example, relaxation editing via pulse sequences like CPMG can be used to detect only small molecules. Ultrafiltration techniques, using appropriately-selected filters, can take advantage of molecular size differences to separate proteins from dissolved metabolites in serum samples before acquiring spectra.

The most noticeable effect of the high protein content on the spectrum of a typical serum sample is baseline distortion. A serum spectrum acquired using a standard NOESY-presaturation pulse sequence is a combination of both protein and metabolite resonances (Figure 1).

A common method of removing the contribution of the broad protein resonances is the use of an NMR-edited pulse sequence such as CPMG, which allows the resonances from larger molecules to relax before detecting the longer-lived resonances from small molecules. These CPMG or "relaxation-edited" spectra have improved baselines (Figure 2), and contributions from weaker signals in the aromatic region are clearly visible (inset, 6.0 to 9.0 ppm, Figure 2).

Relaxation-edited spectra remove protein resonances while retaining the lineshape and position of the residual metabolite peaks. As a result, chemometric approaches have been proven successful when applied to this type of data [18]. Chenomx NMR Suite can prepare your data for this type of analysis using the spectral binning functions in the Profiler module.

Relaxation-edited experiments do not remove the proteins themselves from serum samples, so molecules that bind to proteins can still be visibly affected. For example, the trimethylsilyl group of DSS interacts non-specifically with proteins (Figure 3). The apparent linewidth of the DSS in the serum sample shown is ~15 Hz, much larger than the linewidth for other metabolites such as glucose, which does not interact with proteins (Figure 4). The effects of protein-metabolite interaction on metabolite lineshape varies, ranging from practically unaffected to strongly broadened due to tight binding with the protein.

Before applying targeted profiling to relaxation-edited spectra in Chenomx NMR Suite, the apparent linewidth of the chemical shape indicator (CSI) must be adjusted in the Processor module (Figure 3). The first spectrum will require a trial-and-error approach; in the example shown, the linewidth was adjusted from ~15 Hz to ~1 Hz. The suitability of the final linewidth in this case was determined based on how well the library signature of glucose matched the actual peaks (Figure 5). Subsequent spectra in a dataset will likely require similar adjustments of the linewidth of the CSI peak. Due to this approximation of the CSI linewidth, absolute quantification requires confirming metabolite concentrations externally. Spiking the sample with a known amount of one or more metabolites can provide the necessary reference data. Relaxation-edited spectra can provide relative concentrations by normalizing all measured values to that of one of the native metabolites in the sample. Relative concentrations determined this way can also be a valuable source of data, and do not require any sample spikes.

Spectra may also be edited in the other direction, as diffusion-ordered spectroscopy (DOSY) can filter out the signal from small molecules, leaving only the broad signals from larger molecules like lipids and proteins. These broad signals can be quantified in Chenomx NMR Suite with custom compound signatures created with the Signature Builder module.

An accurate chemical shift reference is required for targeted binning; however, interaction of the CSI with proteins can induce positional shifts as well as linewidth effects (Figure 6). In the example shown, the actual position of the DSS peak is ~0.03 ppm upfield of the true 0.00 ppm position. The required adjustment was determined using the position of other compounds with well established chemical shifts at pH 7.0 (such as lactate and glucose). The default peak positions for compounds in the Chenomx libraries correspond to a pH of 7.0. Thus, the degree of shift occurring in a sample at the same pH can be estimated based on these default positions.

Protein resonances can also be removed from a serum spectrum by removing the proteins themselves from the serum sample. For many studies, lipids, proteins and other macromolecules are not of interest, and may be removed from the serum samples. Elimination of these signals allows analysis of the samples using standard techniques, including NOESY-presaturation pulse sequences (see Analysis Methods). Among the numerous extraction methods available [19], ultrafiltration provides the most reliable quantification, and acetonitrile precipitation provides the best balance between ease of use and number of metabolites observed.

The size difference between macromolecules and small molecule metabolites provides an excellent basis for size-based filtration, or ultrafiltration, of the serum. The easiest and most reliable method is to pass the sample through a molecular weight filter with an appropriate cut-off, separating the proteins from the metabolites.

For most serum samples, a 3 kDa molecular weight cut off (MWCO) micro-centrifuge filter is sufficient to separate metabolites from proteins. The filters should be washed with water at room temperature prior to use, as they are coated with small molecule membrane preservatives such as glycerol, which can interfere with the analysis. Also, adding a rinsing step can improve recovery of metabolites from the protein component. For example, after filtering a 500 μL serum sample, adding 100 μL of D₂O to the protein component and re-centrifuging can almost entirely recover metabolites that are not tightly bound to proteins.

The CPMG spectrum of a serum sample has a very broad DSS peak and a number of baseline distortions near 0.9 and 1.3 ppm (Figure 7), and the lineshapes of the metabolite resonances are distorted due to off-resonance effects from the CPMG pulse sequence. The NOESY spectrum of a sample acquired after molecular weight-based separation is much cleaner, with a better baseline and improved DSS lineshape characteristics (Figure 8). This is important for accurate quantification in Chenomx NMR Suite, although spectra acquired with the CPMG method can also be analyzed with some adaptations, as discussed earlier.

Another option for removing proteins is chemical precipitation using acetonitrile or other organic solvent. Adding the organic solvent to serum and thoroughly mixing makes the protein precipitate out of solution; centrifugation helps to separate the solid out of the sample. This method allows the extraction of metabolites bound to proteins, but absolute quantification may not be consistent from sample to sample. Although precipitation is not as reliable as ultrafiltration, it has some advantages in terms of ease of processing and sensitivity [19].

Anticoagulants such as heparin or citrate are often present in plasma samples, and may interfere with NMR spectral analysis. For example, large amounts of non-deuterated citrate in a plasma sample can saturate the NMR receiver during acquisition. Also, signals overlapping the citrate resonances in such samples may be difficult to identify or quantify. Heparin is a large molecule, so resonances associated with heparin can be removed using relaxation-edited techniques or ultrafiltration as described earlier. The choice of heparin counter-ion is important, as common ions such as lithium, sodium and ammonium can affect the ionic strength of the sample matrix, and thus alter the appearance of the spectrum.

Analyzing NMR spectra of blood serum or plasma requires dealing with the effects of proteins and other large molecules. A variety of methods are available that can reduce the effects of these large molecules on sample spectra, including NMR-edited pulse sequences to remove the problem signals, and ultrafiltration to remove the large molecules themselves. In combination with targeted profiling with Chenomx NMR Suite, these techniques allow accurate identification and quantification of a variety of small molecule metabolites.

More details on the experimental techniques described in this note are available. Please contact us at support@chenomx.com for more information.