EVALUATION OF APPROACHES FOR QUANTITATIVE TARGETED PROFILING OF COMPLEX COMPOSITIONS USING 1D 1H NMR SPECTROSCOPY

Doris M. Jacobs, Ewoud van Velzen, Velitchka Mihaleva

1. INTRODUCTION

Profiling of low molecular weight (LMW) compounds provides an 'unbiased' and broad view on the composition of foods and biofluids and thus is a promising approach to control the quality and safety of foods ' as well as to assess the effect of foods/nutrients in biological systems. One-dimensional (1D) 1H NMR-based profiling of LMW compounds is a common method, as it allows for simultaneously measuring a wide range of compounds with different physico-chemical properties in foods and biological fluids. It is often preferred over other analytical techniques such as LC/MS because of its speed, reproducibility and versatility despite its low sensitivity (approx > 0.1 mg/g). In addition, NMR spectroscopy delivers quantitative information by using internal or external calibrants such as chemical compounds (trimethylsilyl propionate (TSP-d6), 4,4-dimethyl-4-silapentane-1-sulfonic acid (DSS-d6) or maleic acid) or an electronic signal (ERETIC). In particular, recent advances in instrumental performance such as the PULCON (PUlse Length based CONcentration determination) method provide accurate concentrations using external calibrants. Notably, NMR spectroscopy does not require compound-specific external standards for absolute quantification, which often limits MS analysis if these standards are not commercially available.

Although quantitative NMR (qNMR) has been applied for decades mainly for structural elucidation of pure organic compounds, qNMR has only recently gained in importance for mixture analysis due to the higher resolution and sensitivity of the modern NMR instruments. Yet, the quantification of LMW compounds from 1D 1H NMR profiles nowadays still is a major hurdle because of the signal overlap preventing the accurate integration of NMR signals. This signal overlap can be reduced by acquiring two-dimensional (2D) NMR spectra, in which information is spread out to the second dimension. For example, the 2D 1H-13C heteronuclear single quantum correlation (HSQC) NMR experiments have been applied to quantify a number of metabolites in human urine. However, the data acquisition time is longer and the sensitivity of the 2D spectra is lower compared to 1D 1H NMR spectra. Alternatively, solid phase extraction (SPE) or liquid chromatography (LC) methods can be connected prior to NMR data acquisition to isolate specific compounds or compound classes. This method has mainly been applied to clean up samples, to enhance signals for better identification and quantification of compounds. Recently, SPE-NMR has also been applied to generate complementary, less complex sub-profiles from human urine samples. However, the time saving for the analysis of these sub-profiles goes at the expense of more time for sample preparation when compared to conventional NMR-based metabolite profiling of urine.

Spectral fitting of overlaid signals in 1D 1H NMR profiles is a third option and relies on fitting procedures based on certain constraints, prior knowledge, iterative procedures and convergence criteria. It is probably the most promising option, considering that fitting algorithms can in principle be applied in automation for a series of NMR profiles, in this way retaining the advantages of fast NMR data acquisition and minor sample preparation. Several software programs are currently used for spectral fitting such as AMIX (Bruker Biospin GmbH, Germany), Chenomx NMR Suite (v7.1, Chenomx, Inc Edmonton, Canada), PERCH NMR software (PERCH Solutions Ltd., Kuopio, Finland) and Mnova (MestreLab Research S.L.). Herein, we briefly demonstrate the performances of the two software packages Chenomx NMR Suite and PERCH NMR software for the quantification of compounds in food extracts and serum, respectively. Both software programs are well recognized in the field of metabolomics and quality control of natural products. Yet, they differ in fitting algorithms, automation routines, and user-friendliness. In the following we discuss the most relevant features of both software packages and indicate some limitations. We do not intend to give a detailed description of their operation procedures.

2. DISCUSSION

2.1 Quantitative Targeted Profiling Using Chenomx NMR Suite

Chenomx NMR Suite is composed of several modules. The Profiler module is used to identify and quantify compounds in the experimental spectrum. The 1H NMR spectra can be directly imported into the Profiler module or after processing using the Processor module. In the Processor module the spectra are calibrated to the signal of a Chemical Shape Indicator (TSP, DSS or Formate) to determine the concentration of individual compounds. Compounds are identified and quantified using a frequency-dependent (in this case 600-MHz) library of individual spectra, that were acquired at a temperature of 298 K using a noesypr sequence with an acquisition time of 4 s, a mixing time of 0.1 s, a water suppression delay of 0.99 s (at 58 dB), a sweep width of 12 ppm, and a relaxation delay of 1 s. The Chenomx library includes 1H NMR reference models of 302 metabolites which are managed in the Library Manager module. To quantify and identify the compounds, a Lorentzian peak shape simulation of each reference compound is superimposed upon the corresponding signals of the actual experimental spectrum. The linear combination of the selected reference components from the Chenomx library gives rise to the total spectral fit, which can be evaluated with a summation line.

The collection of I Lorentzian reference models in the Chenomx library is actually a series of quantitative reference spectra si with concentrations ci and relative response factors Florini. The determination of Florini for component i is based on the total spectral integral asi relative to the integral of the internal standard IS (as,ISi) with concentration cISi according to equation 1.

[MATHEMATICAL EXPRESSION OMITTED] (1)

Here ri is a (1 × J) row vector of J response factors of component i, ari and ar,ISi represent the integral values of the response factors of component i and internal standard IS. as,ISi and as,ISi are calculated similarly to asi and ari.

The relative response factors ([Florin]i) can be used to predict concentration levels in unknown samples provided that the same experimental conditions were applied. When standardizing against an internal standard, Florini translates the NMR signal intensities into absolute concentrations. In case the experimental conditions are not identical, Florini may be adjusted with a factor φi according to equation 2:

[MATHEMATICAL EXPRESSION OMITTED] (2)

where [??] is the integral value of the fitted component model from the Chenomx library in an unknown sample, [??] is the integral value of the fitted internal standard model in an unknown sample, εi is the residual error and φi is a constant accounting for experimental bias such as water suppression, T1 and T2 relaxation, properties of the fluid (e.g. salt concentrations, buffer strength, etc.). The linear combination of N fitted model spectra [??](1 × J) from the Chenomx library are matched with the experimental spectrum e (1 × J) by minimizing the sum of squared differences according to equation 3:

[MATHEMATICAL EXPRESSION OMITTED] (3)

Here, [??](1 × J) is a single fitted model spectrum from the Chenomx database. In the Chenomx Profiler module the least squares optimization can either be performed in an interactive or an automatic mode.

Chenomx NMR Suite has many assets, which makes this software attractive for many users. In particular, it stands out due to its user-friendliness. The user interface is programmed as 'what you see is what you get' and allows interactive fitting. Thus, this software is of interest especially for non-NMR experts. Moreover, the export functions allow for a convenient export of the concentrations to other formats for further statistical analysis.

Initially, the automatic fit routine provides estimated concentrations of all metabolites identified from the library. However, these estimated concentrations and the identification of metabolites need to be checked by visual inspection and if necessary manually adapted. Practical experience taught that most estimated concentrations need to be corrected. Usually 14 compounds can be manually fit within 2 min per spectrum. Another drawback is that Chenomx NMR Suite does not work in batch mode. Thus, the quantification of compounds in a series of NMR spectra may become a tedious endeavor.

The library comprises a large number of compounds, which is a particular asset considering that the matching of the experimental spectra with the reference spectra generally is the most convenient approach to identify compounds. The compounds in the library are metabolites from biological fluids such as urine and plasma. This implies that the library may be of limited applicability to other complex mixtures. Nevertheless, many compounds present in natural product such as vegetable extracts can still be quantified using the metabolite library. Figure 1 shows an example for a French beans extract in D2O. Even low abundant compounds such as trigonelline can be identified and quantified.

It is notable, that the library can be complemented by other compounds of interest. The Spin Simulator and Compound Builder modules are tools to create simulations and signatures of own quantitative compounds. The use of these models requires some knowledge of NMR theory. They offer a high degree of freedom, which can be useful for the quantification of polysaccharides with barely-known composition. For example, guar gum is a polysaccharide composed of a linear backbone chain of β (1-4)-linked D-mannose residues to which short 1,6-side-branches of α-D-galactose residues are linked. It is a mixture with variable length of the backbone chains and side branches. Similarly, the locust bean gum (LBG) is a high-molecular-weight hydrocolloidal polysaccharide composed of galactose and mannose units combined through (l-6)-glycosidic linkages. The complete NMR signature of these polysaccharides is only partially known. Nevertheless, reference spectra of these polysaccharide standards can be acquired, built in the Compound Builder model and subsequently incorporated into the library. These reference spectra can be used to determine their ratios in products (Figure 2).

It is notable that Chenomx NMR Suite comes with some special features that allow for certain corrections: First of all, the chemical shift position of the reference compound can be adjusted within a certain pH range (pH 4-9). This is helpful for urine analysis knowing that urine is subject to considerable variations in pH values. However, caution is demanded for reference compounds with multi-spin system: Since the chemical shift positions of the individual signals can be adjusted independently from each other, there is no control whether each signal has been shifted to the same pH value. Secondly, differences in line width between the experimental and reference spectra can be adjusted using the 'shim correction' function correcting some types of systematic errors in the experimental spectrum that may have occurred during acquisition.

Furthermore, it is noteworthy that the reference spectra were acquired under certain experimental conditions which may not be appropriate for the experimental spectrum. For example, differences in the irradiation strength for water suppression or relaxation delays may affect the signal intensities thus leading to deviating concentrations. Therefore, it is advisable to check the validity of equation (2) and, if necessary, to correct the relative response factor fi with φ according to the own experimental conditions. With this φ adjustment, a good agreement (slope = 0.94) and linearity (R2 = 0.99) between known concentrations of several compounds given by the initial weight and those determined by Chemomx NMR Suite was found (Figure 3). Moreover, the Profiler module is capable of determining concentrations from overlaid spectral regions by matching the total spectral fit to the line shape of the experimental spectrum. However, the adjustment of the concentrations may be subjective, especially if not all components of the overlaid regions are known.