Low-Field, Benchtop NMR Spectroscopy as a Potential Tool for Point-of-Care Diagnostics of Metabolic Conditions: Validation, Protocols and Computational Models

Abstract

Novel sensing technologies for liquid biopsies offer promising prospects for the early detection of metabolic conditions through omics techniques. Indeed, high-field nuclear magnetic resonance (NMR) facilities are routinely used for metabolomics investigations on a range of biofluids in order to rapidly recognise unusual metabolic patterns in patients suffering from a range of diseases. However, these techniques are restricted by the prohibitively large size and cost of such facilities, suggesting a possible role for smaller, low-field NMR instruments in biofluid analysis. Herein we describe selected biomolecule validation on a low-field benchtop NMR spectrometer (60 MHz), and present an associated protocol for the analysis of biofluids on compact NMR instruments. We successfully detect common markers of diabetic control at low-to-medium concentrations through optimised experiments, including α-glucose (≤2.8 mmol/L) and acetone (25 µmol/L), and additionally in readily accessible biofluids, particularly human urine. We present a combined protocol for the analysis of these biofluids with low-field NMR spectrometers for metabolomics applications, and offer a perspective on the future of this technique appealing to 'point-of-care' applications.

Keywords: benchtop 60 MHz NMR analysis; biomarkers; biomolecules; diabetes; metabolomics; protocol; validation.

Conflict of interest statement

F. Casanova is employed by Magritek GmbH.

Figures

Figure 1

1H nuclear magnetic resonance (NMR) type 2 diabetic urinary profile acquired on a 60 MHz benchtop instrument, highlighting a clearly distinguishable β-Glucose-C1-H resonance (d, ∂ = 4.65 ppm), in addition to the α-Glucose-C1-H one located at ∂ = 5.25 ppm (d), and all further bulk glucose ring structure protons within the 3.19–3.95 ppm chemical shift range for both anomers. Moreover, resonances arsing from a range of further metabolites such as hippurate-CH, indoxyl sulphate-CH, urea-NH2, Cn-CH3/-CH2, creatine-CH3/-CH2, citrate-CH2 (A/B coupling pattern), glutamine-CH2, acetoin-CH3, acetate-CH3, lactate-CH3, N-acetyl storage compound-NHCOCH3, alanine-CH3, isoleucine-CH3 and leucine-CH3 are also visible in this spectrum. Chemical shifts were referenced to internal tetra-deuterated trimethylsilylpropanoate (TSP) (∂ = 0.00 ppm). Abbreviations: 3-d-HB, 3-d-hydroxybutyrate-CH3.

Figure 2

Calibration curve plot of the α-Glucose:TSP (δ = 5.25:0.00 ppm) resonance integral ratio vs. total glucose concentration in phosphate-buffered aqueous solutions (pH 7.00) containing ca. 10% (v/v) 2H2O (red plot). Glucose concentrations ranged from 10.0–600.0 mmol/L, and that of the TSP internal standard was maintained at a final concentration of 223 µmol/L. The blue plot represents that derived from TSP-normalised integral values predicted directly from the known concentrations of total glucose and TSP present, the relative numbers of 1H nuclei contributing towards their 1H NMR resonances (1 and 9 respectively), and the 36% abundance of the α-glucose anomer (δ = 5.25 ppm signal).

Figure 3

2D 1H-1H COSY NMR diabetic urinary profile acquired on a 60 MHz instrument highlighting connectivities between α-Glucose-C1-H and -C2-H resonances at ∂ = 5.25 ppm (d) and 3.52 ppm (dd), respectively (labelled in red), and correspondingly, those of the β-anomer at δ = 4.65 (d) and 3.23 ppm (dd), respectively (labelled in green. A further 1H-1H COSY connectivity between signals located at δ = 3.06 and 4.08 ppm is also clearly visible.

Figure 4

Plots of mean ± 95% CIs urinary glucose concentrations determined from analysis by low field (LF) 60 MHz benchtop 1H NMR, conventional HF 400 1H MHz NMR, the GOD-PAP spectrophotometric (abbreviated Spectro) and chromophoric dipstick (abbreviated Dipstick) test systems. The wide confidence intervals are predominantly ascribable to the highly statistically significant ‘Between-Participants’ random effect component-of-variance (Pj in Equation (1)), and not analytical reproducibility.

Figure 5

Principal component analysis (PCA) scores plot of PC2 (17.04% of total variance) versus PC1 (64.94% of total variance) for a preliminary investigation of distinctions between healthy control and type 2 diabetic cohorts, and also potential sample outliers. Colour codings: blue, urine samples collected from healthy controls; green, those from type 2 diabetes participants. The black points represent scores plot centroids for the two groups explored. PCA was performed using XLSTAT2014 software, and the dataset was TSP-normalised, generalised logarithmically (glog)-transfomed and Pareto-scaled prior to analysis.

Figure 6

(a) Orthogonal projections to latent structures- discriminatory analysis (OPLS-DA) scores plot of orthogonal T score vs. T score for the TSP-normalised dataset demonstrating a clear metabolomics-based distinction between type 2 diabetic patients and healthy controls. 95% confidence ellipsoids are also shown (the type 2 diabetic patient cluster sample T score values (+2.5 to +6) are all greater than those of the control cohort (−6 to +1)); (b), as (a), but for the constant sum-normalised (CSN) dataset.

Figure 7

(a) Receiver operating characteristic (ROC) curve (plot of true positive vs. false positive rates) with an area under ROC curve (AUROC) value of 0.975 obtained from the support vector machine (SVM) model building system explored with 10 out of a possible 27 variables. ROC curves were developed via Monte Carlo Cross-Validation (MCCV) involving a balanced sub-sampling processes involving application of an SVM model builder (TSP-normalised urinary dataset). The inset shows mean AUROC values estimated for increasing sampling sizes, together with 95% CIs for these values. (b) Probability view arising from a balanced sub-sampling approach for SVM model training (predicted class probabilities for each sample employed the most effective area under ROC curve (AUROC)-based classification system).

Figure 8

Diabetic urinary 1H NMR profile acquired at 60 (red) and 400 MHz (blue) operating frequencies. Assignments: 1, TSP-Si(CH3)3; 2, Citrate-A/B-CH2CO2−; 3, Cn/Creatine > N-CH3; 4, β-Glucose-C2-H; 5, α- and β-Glucose-C4-H/C5-H, and α-Glucose-C2-H; 6, α- and β-Glucose-C3-H/C5-H/C6-H2; 7, Cn-CH2; 8, β-Glucose-C1-H; 9, H2O/HOD-OH; 10, α-Glucose-C1-H.

Figure 9

Proposed schematic representation of biofluid analysis by LF benchtop NMR spectroscopy. Numbers refer to sections of the protocol in Part 5 of this work, relevant to each respective section of the scheme. KEGG: Kyoto Encyclopedia of Gene and Genomics