Very First Application of Compact Benchtop NMR Spectrometers to Complex Biofluid Analysis and Metabolite Tracking for Future Metabolomics Studies: A Retrospective Decennial Report from November 2014

Abstract

Herein we report the very first experiments which were conducted in an attempt to demonstrate the ability of low-field (LF), compact benchtop NMR spectrometers to provide spectral profiles of whole human biofluids, which took place in September–November 2014, and this paper represents a 10-year (decennial) anniversary of this work. LF ¹H NMR analysis was performed on ²H₂O-reconstituted lyophilizates of urine samples (pH 7.00) collected from untreated Niemann-Pick type C1 (NPC1) disease patients and their heterozygous carrier controls (n = 3 in each case). ¹H NMR spectra were acquired on a 60 MHz Oxford Instruments Pulsar compact benchtop spectrometer with spectral filter widths of 5000 Hz, using 1000–1600 scans, and relaxation delays of 15 or 30 s. Further, 400 MHz spectra were also obtained on these samples. Following parameter optimisation, the benchtop system generated reasonable quality urinary ¹H NMR profiles containing ca. 30 signals. Benchtop ¹H NMR analysis confirmed the abnormal urinary metabolic signature of NPC1 disease, and also revealed a gastric permeability disorder in one patient (detection of upregulated urinary sucrose, verified by 400 MHz NMR analysis). Early LF NMR experiments also demonstrated that glucose was trackable in control urine samples pre-spiked with this metabolite. This paper continues with further developments made on LF NMR-based metabolomics technologies, which are systematically discussed for related investigations conducted since 2014. In conclusion, such ‘first-time’ bioanalytical information regarding spectral quality served to pave the way forward for benchtop NMR-based metabolomics investigations of biofluids, which could provide invaluable disease-engendered ‘snapshots’ of disturbances to metabolic pathways and activities, along with those of any co-linked or unlinked comorbidities.

1. Introduction

In this investigation, we retrospectively report the very first experiments targeted on the analysis of human biofluids using a compact, near-portable low-field (LF) benchtop NMR facility, which took place at Oxford Instruments’ then headquarters based in Abingdon, Oxfordshire in September–November 2014, and at the time of submission, this report marks just over the 10th anniversary (decennial) of this novel event. This technology was very newly established at that point in time, and our earliest studies involving LF NMR analysis of biofluids were predominantly focused on the acquisition of at least reasonable quality and reliable and reproducible spectral profiles, and not on the prior provision of any valuable supporting clinical diagnostic information. However, we were quite surprised at the results acquired, and from these experiments we felt that we had certainly swung open a number of research doors regarding the employment of this then very new technology for the diagnosis and potential prognostic monitoring of human and animal diseases.

Since that time, this device, its upgrade and related spectrometers for benchtop compact NMR analysis have been increasingly employed for the purpose of evaluating the metabolic profiles of a now rapidly expanding series of human and animal biofluids, experiments which are predominantly centred on the applications of this novel technology for the rapid diagnosis of a series of diseases. Indeed, these experiments have further stretched to the coupling of this form of spectroscopic analysis to the application of modern metabolomics techniques, i.e., through the multivariate (MV) identification of characteristic ‘patterns’ of biomolecules which may serve as useful biomarkers.

Investigations more recently performed by our research group have already established that this non-stationary LF NMR facility can offer much valuable diagnostic potential regarding the multicomponent analysis of biofluids, and since 2018–2019 we have achieved a full series of peer-reviewed reputable research publications on the clinical/disease diagnostic applications of such benchtop NMR facilities [1,2,3,4,5,6,7]. Indeed, preliminary pioneering results acquired have demonstrated the identification of up to 15–20 biomolecules simultaneously in human urine or saliva samples at this low operating frequency, along with the quantification of a smaller number of those with clearly visible resonances, at least some of which are viewed to be largely unaffected by potential interferences arising from superimposing, albeit lower intensity, adjacent signals. Indeed, the routine diagnosis of both type 1 and 2 diabetes (T1D and T2D, respectively) is both rapid and simple with this LF device, and this strategy often uses more than several biomarkers for this purpose, especially when they are associated with disease comorbidities, which may also be trackable [4].

Here, we retrospectively report the very first successful application of a low-field (LF) compact benchtop NMR facility to explore its ability to acquire reliable and reproducible ¹H NMR spectra on human biofluids, with urine being used as a target analyte matrix. For this purpose, we utilised and optimised a then early-stage 60 MHz Oxford Instruments Pulsar compact benchtop spectrometer in 2014, and for the first time we show the first example of the rewarding employment of this facility to investigate the chemopathological basis of human disease morphology, and an associated gastrointestinal mobility comorbidity, in a rare debilitating lysosomal storage disease (Niemann-Pick type C1 (NPC1) disease). This early pioneering study also paved the way forward for future experiments which later demonstrated the abilities of LF benchtop NMR devices to track urinary glucose and related biomarker concentrations in subjects with diabetes conditions [1,3,4]. In accordance with this observation, this decennial report is supported and supplemented by additional more recent (post-2014) results obtained from some of our corresponding follow-up studies (including those involving the identification and monitoring of patients with T2D from their urinary ¹H NMR profiles [1,2,3,4]), along with those led by other laboratories worldwide.

2. Materials and Methods

2.1. Urine Sample Collection

Urine samples (n = 3) analysed in this compact NMR analysis investigation were collected from a UK clinical cohort of untreated NPC1 patients (age range 9–28 years, 1 male/2 female) and n = 3 corresponding disease-free parental heterozygous carrier controls (age range 28–52 years, 1 male/2 female). These UK-based studies were collected with informed consent and approved by the appropriate Research Ethics Committee (12/NW/0753).

Prior to commencing the study, all participants (and/or their legal guardians, where required) were provided with a participant information sheet (PIS), and after reading this and considering their participation in the study, were then required to sign a research project consent form in the presence of a suitable staff member witness. The PIS clearly informed those recruited that in view of their voluntary participation, they were able to withdraw from the study at any stage of its duration. Participants were also requested to completely refrain from the consumption of any alcoholic beverages, and any dietary sources of this agent, for 24 h. before the collection of urine samples. All ethics considerations were in accordance with those of the Declaration of Helsinki of 1975 (revised in 1983).

As noted in Section 2.3 below, the above control samples were also treated with D-glucose (Merck KGaA, Darmstadt, Germany), plus further metabolites, in ‘spiking’ experiments originally designed to confirm resonance assignments (final added glucose level 50–200 mmol/L). These particular urine samples also served as experimental models for mimicking those obtained from T2D patients with a poor level of glycaemic control. However, analytical samples generated therefrom in the 2014 experiments were very restricted in view of the limited volumes of biofluid available for them.

2.2. Urine Sample Storage, Processing and Preparation

Human urine samples collected were anonymised using a computer-generated randomised sample code, and then routinely stored at −80 °C. When ready for analysis, measured 2.0–3.0 mL aliquots of these samples were routinely thawed at ambient temperature and centrifuged at 4 °C to remove any cells and debris (5000 rpm for a period of 10 min.), and the clear supernatants were refrozen and then lyophilised. The supernatant residue was dissolved in 0.60 mL of a ²H₂O solution containing 20.00 mmol./L phosphate buffer (pH 7.00) and 0.04% (w/v) sodium azide, the former to maintain sample pH, the latter to protect against microbial infiltration following sample collection, preparation and storage. These solutions also contained 275 µmol/L TSP as a δ = 0.00 ppm ¹H NMR chemical shift reference and quantitative internal standard. Since at that time no water signal presaturation/quenching technologies were available for the LF NMR device employed, as with any others available then, reconstitution of the samples in ²H₂O was necessary; this deuterated NMR solvent also served as a field frequency lock.

2.3. LF ¹H NMR Measurements at an Operating Frequency of 60 MHz

From NPC1 disease patients, 60 MHz ¹H NMR spectra of ²H₂O-reconstituted lyophilizates derived from urine specimens were collected, and their heterozygous carrier control subjects were acquired on a 60 MHz Oxford Instruments Pulsar compact benchtop spectrometer for a spectral filter width of 5000 Hz, with 1000 or 1600 scans, and a relaxation delay of 15 or 30 s. These 60 MHz spectra were processed by zero-filling to 128 K datapoints, the application of apodization (exponential −1 Hz, Gaussian +1 Hz), manual phasing and baseline correction. Chemical shifts in resonances present in all spectra acquired were referenced to the −Si(CH₃)₃ signal of added sodium 3-trimethylsilyl [2,2,3,3-²H4] propionate (TSP) at δ = 0.00 ppm (Fisher Scientific Ltd., Loughborough, UK).

Continuing ‘metabolite assignment’ experiments coordinated by our group generated LF benchtop NMR spectra on NPC1 heterozygous carrier control (non-diabetic) urine samples which were originally pre-spiked with relatively high levels of glucose (50.0–200.0 mmol/L); as with other urine samples, these were acquired on buffered ²H₂O-reconstituted lyophilizates, in accordance with experiments originally performed. These samples were also used to serve as ‘mimics’ for those collected from T1D or T2D subjects with poor glycaemic control, albeit also without any associated biomarkers such as ketone bodies present. With the use of calibration curves, this biomolecule was quantifiable from the intensity of its δ = 5.25 ppm −C1H α-anomeric proton doublet, and normalisation of this parameter to TSP’s −Si(CH₃)₃ δ = 0.00 ppm singlet resonance after allowing for differential numerical contributions of respective ¹H nuclei towards these. Such determinations were brought about using the application of an iterative least-squares fit deconvolution model using Lorentzian/Gaussian line shapes (ILSFDM), as described in Ref. [8]. ¹H NMR resonance bucketing was conducted manually.

Although not strictly necessary because spectrometer software water resonance presaturation techniques were not applicable, nor available then, we also employed a series of linear D-glucose calibrant standard solutions containing the same level of the TSP internal standard, so that glucose concentrations could be semi-quantitatively determined from datapoint-limited plots of its normalised signal intensity versus its concentration (r = 0.93–0.98). The 36:64% ratio of the α:β anomeric form ratio of glucose was also accounted for in these estimations.

Signal-to-Noise (STN) Ratio Values

Given the complex baseline present across the spectrum, the NMR line heights were taken from the top of the lines to a software-fitted slope across their visible bases. Absolute vertical heights from the baseline were not employed, and therefore the STN ratio values presented remain somewhat conservative. The basic formula for calculating the STN ratio was: STN ratio = Resonance Line Height/Noise Amplitude, both parameters being measured in pixels. Baseline STN values for all signal-free 10.00–12.00 ppm spectral regions were 1.0 pixel.

2.4. MF NMR Measurements (Operating Frequency 400 MHz)

Corresponding medium-field (MF) spectra were also acquired on these anonymised samples utilising a 400 MHz Bruker Avance I (AVI) NMR spectrometer (Leicester School of Pharmacy, De Montfort University, Leicester, UK) operating at a frequency of 399.93 MHz; these samples were not lyophilized, nor ²H₂O-regenerated in this case. All urine specimens were analysed using the noesygppr1D water suppression pulse sequence in order to presaturate the intense water signal (δ = 4.80 ppm) in spectra obtained, with irradiation of the residual water signal at this frequency occurring during the recycle and mixing time delays. 32 K data points were acquired using 128 scans and 2 dummy scans, and 3 µs pulses, throughout a sweep width of 4844 Hz, a receiver gain of 128 and a probe temperature of 298 K. Spectra were acquired in an automated manner using a sample changer for continuous sample delivery. This auto-sampler facility ensured that the samples were continuously and randomly delivered throughout the analysis duration. Exponential line-broadening functions of 0.30 Hz were routinely applied to the FIDs prior to Fourier transformation, and all spectra were manually phased and baseline-corrected. Corresponding 400 MHz noesygppr1D ¹H NMR spectra of ca. 5–8 mmol/L of solutions of sucrose, and other single model urinary metabolites in 50 mmol/L phosphate buffer (pH 7.00) were similarly acquired.

As with the 60 and MHz spectra, chemical shift values were referenced to the −Si(CH₃)₃ resonance of added TSP (δ = 0.00 ppm) at 400 MHz operating frequency. Morover, the −CH₃ group signals of acetate (s, δ = 1.920 ppm), alanine (d, δ = 1.487 ppm), lactate (d, δ = 1.330 ppm) and creatinine (>N-CH₃ s, δ = 3.030 ppm) served as secondary internal references, where detectable.

At this operating frequency, calibration samples containing 0.00–20.00 mmol/L added sucrose and 0.28 mmol/L TSP internal standard were prepared, and plots of the ratio of the electronic intensity of the former’s -C1H proton signal (δ = 5.40 ppm) to that of the latter internal standard (δ = 0.00 ppm) were linear (r = 0.9942). From these plots, the concentration of urinary sucrose was determined, although it was not necessary to correct for the adverse attenuating influence of the water resonance presaturation process employed on sucrose’s -C1H signal intensity, since each calibration standard solution underwent the same noesygppr1D presaturation strategy during acquisition.

2.5. Assignment of ¹H NMR Resonances of Human Urine Samples

The identities of biomolecule resonances present in urinary spectra acquired were routinely assigned by a consideration of chemical shift values, coupling patterns and coupling constants (and also from a range of literature sources), and then cross-checked with the Human Metabolome Database (HMDB) [9], including those for urinary sucrose and glucose. As usual with our metabolomics experimental protocol, a combination of both one- (1D) and two-dimensional (2D) ¹H-¹H correlation (COSY) spectroscopic techniques were employed to confirm these assignments (the latter for the 400 MHz spectrometer alone), as was the ‘spiking’ of these biofluids with appropriate small µL volumes of ca. 6–8 mmol/L standard solutions of authentic biomolecules prepared in 50 mmol/L phosphate buffer (pH 7.00), as noted above.

3. Results

3.1. Very First LF 60 MHz ¹H NMR Spectra of Human Urine

Following the implementation of a number of stages for the optimisation of spectral acquisition on the benchtop NMR system employed, and selective control and operation of the parameters outlined above, the authors were successful in attaining at least some reasonable ¹H NMR spectral profiles on human urine samples. However, the best quality spectra with the highest signal-to-noise (STN) values were acquired for many hours over the weekend using >1000 scans and a 30 s relaxation delay. Historically, Figure 1 shows the very first 60 MHz ¹H NMR spectrum of a urine sample collected from a patient with NPC1 disease in November 2014, which contained approximately 30 visible signals arising from a series of biomolecules therein, some of which are known to serve as diagnostic markers [10]. However, as expected at this operating frequency, although at least some of these resonances appeared to be free or relatively free of any interfering spectral overlap problems arising from neighbouring ¹H NMR signals, those which were potentially quantifiable were limited somewhat by this restrictive superimposition, most especially at their date of acquisition in late 2014.

Table 1. Detailed list of resonances observable in the very first LF compact ¹H NMR spectra acquired on human urine samples at an operating frequency of 60 MHz, along with their assignments. The biomolecules detectable correspond to those visible in the sample collected from an NPC1 disease patient shown in Figure 1. The superimposition status shown in the final column refers to the resonance overlap phenomena experienced by each ¹H NMR signal indicated, with +, ++ and +++ referring to low, medium and high levels of superimposition potential from neighbouring ¹H NMR resonances, and ni indicating no significant interference observed. Abbreviations: BCAAs, branched-chain amino acids; GlycA, ‘glycoprotein A’ resonance. * Indicates tentative assignments made at 60 MHz operating frequency, but not at 400 MHz. ^¶ Indicates that in view of volatility and/or decomposition, signals of these metabolites are not readily visible on analysis of samples which were pre-lyophilized, a necessary step for analysis using early LF 60 MHz spectrometers, although they were with analyses on later 60 MHz benchtop, MF and HF NMR instruments equipped with water signal presaturation software.

Assignment	Assignment Code	Chemical Shift (δ) Value (ppm)	Corresponding Coupling Pattern(s)	Superimposition Status
BCAA-CH3s (isoleucine/leucine/valine)	1	0.86–1.05	d/t/dd	+++
3-Aminoisobutyrate-CH3	2	1.15	d	+
3-D-Hydroxybutyrate-CH3	2A	1.24	d	+
Lactate-CH3	3	1.33	d	+
Alanine-CH3	3A	1.49	d	ni
Acetate-CH3	4	1.92	s	ni
GlycA/N-Acetylsugar- and N-Acetylamino acid-NH-CO-CH3	5	2.04	s (broad)	++
¶ Acetone-CH3/* 5-Aminovalerate-α-CH2	5A	2.24	s/2 × m	+++
¶ Acetoacetate-CH3/Aminoadipate-γ-CH2	5B	2.28/2.32–2.35	s/2 × dt	++
Pyruvate-CH3 and Succinate-CH2s	6	2.38–2.41	s and s	+++
Glutamate-/Glutamine-γ-CH2	6A	2.34–2.43	m/m	+++
Citrate-CH2-CO2−	7	2.60	Apparent d	ni
Dimethylamine-N(CH3)2	8	2.75	s	ni
Dimethylglycine-N(CH3)2	9	2.83	s	+
Creatine-/Creatinine-N-CH3	10	3.03	2 × s	+++
* Malonate-CH2	10A	3.11	s	++
* Dimethylsulphone	10B	3.17	s	+++
Choline/Betaine-N+(CH3)3	11	3.20	s/s	+++
Trimethylamine N-oxide-N(CH3)3	12	3.25	s	++
Methanol-CH3	12A	3.38	s	+
* Taurine-CH2SO3−	12B	3.46	t	++
Glycine-CH2/Sucrose sugar ring-Gluc-C2H	13	3.56/3.55	s/dd	++
Amino acid-α-CH protons	14	3.7–3.8	All m	+++
Creatine-CH2/* Glycolate-CH2/Hippurate-CH2	14A	3.95	s/s/d	+++
Sucrose sugar ring-Fruc-C1′(-CH2OH) and Gluc-C3H/* Phenylacetylglycine-CH2	S1	3.67 and 3.75/3.66	s/t/s	+++
Sucrose sugar ring-Fruc-C5′H, Gluc-C5H, Fruc-C6′(-CH2OH) and Gluc-C6(-CH2OH)/* Ethanolamine-CH2OH/* Mannitol-CH2OH	S2	3.89/3.87/3.82/3.82/3.84	dd/dd/m/m/t/dd	+++
Sucrose sugar ring-Fruc-C4′H/Creatinine-CH2	S3	4.04/4.05	t/s	++
Sucrose sugar ring-Fruc-C3′H	S4	4.21	d	+
Sucrose: D-glucopyranoside-C1αH	S5	5.40	d	ni
Urea-CONH2	15	5.75	Broad s	+
* Tyrosine aromatic ring-C2H,C6H/* Histidine imidazole ring- C4H/Phenol-C2H,C4H/* 4-Hydroxypropionate-C2H,C6H and C3H,C5H/* 3-(3-Hydroxyphenyl) propionate-C2H,C4H	16	6.80–7.12	d/s/2 × d/2 × m	+++
Indoxylsulphate-C5H, C6H/* Imidazole ring protons-C2H,C4H,C5H/* Tyrosine-aromatic ring-C3H,C5H	17	7.21–7.28	2 × dd/s/d	+++
Indoxylsulphate-C2H/* Phenylalanine- aromatic ring-C3H,C4H,C5H	18	7.32–7.44	s/m	+++
Phenylalanine aromatic ring-C2H,C6H	19	7.30–7.33	m	++
Indoxylsulphate-C2H	20	7.40	s	++
Indoxylsulphate-C7H/Hippurate-aromatic ring-C3H/C5H	21	7.51–7.54	d/m	+++
Hippurate-aromatic ring-C4H	22	7.63	tt	+++
Indoxylsulphate-indole ring-C4H	23	7.77	d	++
Hippurate-aromatic ring-C2H,C6H	24	7.81	dd	+
Formate-H	25	8.46	s	ni

provides a list of resonances visible in LF 60 MHz spectra acquired on human urine samples, the abbreviations +, ++ and +++ indicating increasing degrees of superimposition with adjacent ¹H NMR signals, which correspond to deceasing potentials for corresponding metabolite quantification (ni represents no significant overlap notable). Nevertheless, clear, mainly interference-free signals were visible for the alanine-CH₃, acetate-CH₃, and formate-H protons, and those of the citrate-CH₂ and dimethylamine-N(CH₃)₂ groups; therefore, even with this primordial stage compact biofluid analysis NMR set-up available then, in principle it was suggested that it would be possible to quantitate the levels of these urinary biomolecules with little complication via the expression of their resonance intensities relative to that of the TSP internal standard (after allowing for their differing numbers of proton contributors). Moreover, for the + superimposition status resonances shown, specifically those of 3-aminoisobutyrate-, 3-D-hydroxybutyrate-, lactate-, dimethylglycine- and methanol-CH₃ groups, urea’s -CONH₂ and hippurate’s-C2H,C6H aromatic ring protons, it may also have been possible to integrate these signals and quantify their biomolecular assignments, although clearly much caution should be applied before attempting this. Furthermore, visible additional resonances labelled with superimposition status scores of ++ and +++ were at that time ruled out of those which classified as quantifiable, although those which from HF spectra were known to be of relatively high intensity in the ¹H NMR profiles of human urine, e.g., sometimes the broad GlycA glycoprotein signal located at δ = 2.04 ppm, contained only small contributions arising from adjacent signal interferants (possibly <10% of the total δ = 2.00–2.14 ppm signal), and therefore its quantification as N-acetylsugar GlycA equivalents may have also been possible, assuming that detectable concentrations of low-molecular-mass N-acetylsugars such as N-acetylglucosamine or N-acetylneuraminate as sharper signals are largely absent.

Moreover, where present, doublet resonances assignable to the -C1H proton regions of carbohydrate species were sometimes detectable, and for this study distinctions between urinary sucrose (δ = 5.40 ppm) and α-glucose (δ = 5.25 ppm) were achievable. Indeed, J values for these resonances were 3.8 and 2.7 Hz, respectively, and from the centralised span of these doublets, these were virtually resolvable at an operating frequency of only 60 MHz, although use of the ILSFDM deconvolution software facilitated their quantification in such situations.

Figures S1.1–S1.3 (Supplementary Materials Section S1) show further examples of typical 60 MHz benchtop NMR spectra acquired on ²H₂O-regenerated urinary lyophilizates during this study. These include those from a further NPC1 patient and a heterozygous carrier control subject, together with one from a control sample which was pre-treated with a high concentration of D-glucose in order to simulate that derived from a T2D patient with poor glycaemic control.

Some typical STN ratio values for a number of urinary biomolecule resonances detectable at an operating frequency of 60 MHz in urinary lyophilizates are provided in Section S1 of the Supplementary Materials paper accompaniment.

Although not typically applied to benchtop NMR spectroscopy conducted in this study, 2D NMR analysis is a powerful and valuable strategy to use for resolving overlapping resonances, along with the verification of biomolecular assignments. It is also of much utility for the structural elucidation and identification of new or previously unassigned molecules [11,12]. However, the 2D ¹H-¹H COSY technique was applied in order to confirm some assignments made at the 400 MHz operating frequency. This 2D NMR approach segregates resonances on the basis of their chemical shift and J-coupling constant parameters, and this markedly promotes the acquisition of molecular structural and quantitative data, rendering it highly advantageous for the analysis of complex biochemical/chemical admixtures, such as those of human biofluids. However, the long periods required for preparative and acquisitional protocols gives rise to major time-lengths for its operation. To circumvent this limitation, however, researchers are currently evaluating optimised computational and machine-learning strategies [11,12]. However, although the ultrafast 2D NMR technique has shown some promise in various research areas, to date its value in LF benchtop NMR analysis remains to be fully ascertained [13], especially for biofluids. Likewise, the heteronuclear 2D HSQC analysis featuring coupled ¹H and ¹³C nuclei may also be usefully employed for structural elucidations and verifications, and this has proven to be particularly valuable for biofluid analysis, e.g., [14]. Nevertheless, although also applicable to benchtop instruments, the concentrations of analytes required for it to achieve sufficient sensitivity are restrictively high, ca. >100 mmol/L [15].

3.2. Prima Facie Benchtop ¹H NMR Analysis of Urine Collected from Patients with NPC1 Disease and Their Heterozygous Carrier Controls

Lysosomal storage diseases (LSDs) represent a series of inherited metabolic diseases which are delineated by the build-up of undigested macromolecules in lysosomes, a phenomenon which is attributable to enzyme deficiencies in the lysosome [16].

Unexpectedly, this very first LF benchtop NMR spectrum was acquired on a urine sample from a patient with the LSD NPC1 disease, and was found to contain a series of quite intense multiplets within the 3.4–4.3 ppm chemical shift range, signals which are not usually visible or as prominent in corresponding spectra collected from healthy humans, nor those of heterozygous NPC1 carrier controls. Computational checks, coupled with a full review of chemical shift values, coupling patterns and coupling constants, together with the acquisition of this spectrum at a higher operating frequency (400 MHz) as detailed below, revealed that these coupled signals all arose from the carbohydrate sucrose. Indeed, the identity of this disaccharide was readily verified through visualisation of its glucopyranoside-C1αH proton signal (δ = 5.40 ppm), along with those of its characteristic sugar ring protons located in the δ = 3.5–4.3 ppm spectral range, which are also listed in Table 1. The majority of these signals were readily assignable, and these assignments were clearly verified via an examination of the 3.00–5.60 ppm region of 400 MHz spectra acquired on the same sample (Figure 2).

LF NMR analysis of two further NPC1 patient urine specimens, along with those from n = 3 (parental) heterozygous carrier controls, did not reveal sucrose as a metabolite, although it should be noted that its δ = 5.40 ppm glucopyranoside-C1αH doublet resonance was visible in the corresponding 400 MHz spectrum acquired on one of the other NPC1 disease samples analysed, although this was present at a much smaller intensity (and hence concentration) level. Additionally, the NPC1 disease-induced metabolic disturbances previously observed in this group of participants were indeed also visible at 60 MHz, although in view of the low operating frequency of the device applied, these were largely limited to relatively large urinary enhancements in 3-AIB concentration (Figure 1 and Figure 2). The high concentrations of sucrose detectable in the first of the three NPC1 disease samples analysed at 60 MHz reflects the development of a gastric permeability comorbidity issue in one of the patients investigated here, and this was not at all known to the researchers prior to conducting benchtop NMR analysis of it. The mean ± repeat analytical SD value for sucrose estimation on this sample was 2.88 ± 0.17 mmol/L (n = 3).

In addition to the absence of any ¹H NMR-detectable sucrose, LF benchtop spectra acquired on all urine specimens collected from the heterozygous carrier control sampling group did not show any notable metabolic upregulations in 3-AIB, and these results, along with those obtained on a number of other biomarkers, were consistent with the metabolic changes found in our metabolomics study conducted at an NMR operating frequency of 400 MHz [10].

As noted in Section 2.3, NPC1 disease heterozygotic carrier control urine samples treated with selected levels of biomolecules according to our approved protocol served to facilitate the identification of analytes present in this biofluid collected from NPC1 disease patients. However, those ‘spiked’ with D-glucose served as valuable assets for diabetes diagnosis in authentic samples, since they acted as suitable ‘surrogate’ biofluids for patients with uncontrolled or poorly controlled T1D or T2D, and in which glucose acts as the only biomarker, i.e., as a highly upregulated metabolite (Figure S1.3).

4. Discussion of Early LF Benchtop NMR Experiments Applied to NPC1 Disease Urine

Compact LF ¹H NMR Monitoring of LSDs and Associated Gastric Permeability Complications in Human Urine

A full outline of the ¹H NMR-linked metabolomics analysis of human urine collected from NPC1 patients performed at an operating frequency of 600 MHz is provided in Ref. [8]. This study compared the urinary metabolic profiles of young patients afflicted with this disorder to those of their parental heterozygous controls, and discovered significant upregulations in the urinary excretions of bile acids, branched-chain amino acids (BCAAs) and their intermediates and degradation products, including 3-AIB, glutamine, 3-methylhistidine, creatine and succinate in NPC1 patients. Correlated component regression (CCR) and genetic algorithm (GA) models employed for MV data analysis were both found to achieve excellent classification successes, the latter at rates of 96–99%.

In the very first 60 MHz compact ¹H NMR spectra acquired in the current paper, the detection of relatively high levels of urinary sucrose in one of the three NPC1 patient samples was completely unexpected. However, the urinary excretion of high levels of sucrose is characteristic of a gastric permeability complication with the NPC1 patient under study. Indeed, such carbohydrates have previously been employed as markers for gastrointestinal (GI) damage, since unlike damaged mucosa, healthy GI tissue is virtually totally impermeable to disaccharide species [17]. Originally, employment of the disaccharide sucrose as a ‘mild’ upper GI damage marker was put forward by Sutherland et al. [18]; this sugar is degraded within the small intestine by an explicit disaccharidase enzyme. Hence, even with intestinal disorders, the detection of sucrose in both urine or blood following its oral ingestion (100 g) reflects gastric permeability complications, without any interferences exerted by intestinal disorders. Using a threshold cut-off value of 180 mg sucrose per 5 h. urine sample (equivalent to a concentration of ca. 1.80 mmol/L on the basis of a mean 24 h. human urinary excretion volume of 1400 mL), this testing system has offered a sensitivity of 84% for gastric ulcer detection, along with a specificity of as high as 96% for normal endoscopy [18]. This threshold concentration of sucrose is, of course, readily detectable and quantifiable by medium- and high-resolution ¹H NMR analysis, and for our NPC1 disease sample, it was found to be 2.88 ± 0.17 mmol/L when estimated from its non-C1H carbohydrate ring proton signal intensities in spectra obtained at an operating frequency of 400 MHz. Moreover, related animal studies have verified that sucrose served as an ideal ‘site-specific’ marker of gastroduodenal permeability, and that the permeability of this sugar is clearly enhanced by gastroduodenal damage induced by the administration of ethanol or non-steroidal anti-inflammatory drugs such as aspirin [19].

LSDs can affect various organs and biosystems, including the GI tract, and therefore the development of gastric permeability problems is one complication of NPC1 disease. This adverse development may be ascribable to (1) the deleterious accumulation of undigested lysosomal substrates, which may impair normal cellular functions, inclusive of those within the GI tract; (2) chronic inflammation, which may result from lysosomal deterioration; and (3) the malabsorption of nutrients, together with additional GI issues, which can also be induced by lysosomal dysfunctions [20,21,22]. Such GI defects usually manifest themselves as symptoms, including diarrhoea, malabsorption and abdominal pain, the management of which usually involves a holistic strategy featuring dietary adjustments and enzyme replacement treatments, along with other recommended therapies [23]. Further information on these phenomena is available in Refs. [24,25].

Moreover, metabolites from amino acid metabolism pathways were readily detectable, including creatine, alanine and glutamine (from arginine and proline/glycine, serine and threonine, and alanine and glutamine metabolism, respectively); N-acetylsugars from amino-sugar and further amino acid metabolism; succinate from the citric acid cycle; pyruvate from pyruvate metabolism; and the co-metabolite trimethylamine (TMA) from the microbial metabolism of gut flora. Additionally, of especial interest to this study is the ready detection of 3-aminoisobutyrate (3-AIB) [10], which arises from the degradation of branched-chain amino acids (BCAAs), and which is also ¹H NMR-detectable at this low operating frequency, as shown in Figure 1, although it may also be biosynthesised via pyrimidine (thymine) catabolism, a route which also involves L-glutamine. This metabolite is very significantly upregulated in NPC1 patients over those of heterozygous controls. The low-molecular-mass N-acetylsugar and/or broader GlycA resonances have also been reported to be upregulated in NPC1 patients over those of healthy controls [10].

However, it should also be noted that single nucleotide polymorphisms (SNPs) in the AGXT2 (D-3 aminoisobutyrate-pyruvate aminotransferase) gene serve as major causalities of significantly increased urinary excretions of its 3-AIB substrate, i.e., β-aminoisobutyric aciduria, which is commonly found in Asian populations [26]. Indeed, ca. 40% of these populations are high excreters of this metabolite. Since a significant number of the NPC1 patients in the UK population originally recruited to this study were of Asian descent [10], at least some of the elevated urinary 3-AIB concentrations found in this study may also be partially explicable by this aetiology.

5. Developments, Updates and Future Perspectives of the Metabolomics Applications and Potential of LF Benchtop NMR Analysis

5.1. Benchtop NMR-Based Metabolomics Analysis of Type 2 Diabetes (T2D) in Humans

In one of our more recent investigations focused on the diagnosis and potential prognostic monitoring of T2D in humans [4], metabolomics analysis of ¹H NMR datasets acquired on a compact 60 MHz spectrometer, which benefitted from a PC-installed software-driven water signal presaturation advantage, revealed clear distinctions between metabolic clusterings of the T2D and healthy age-matched control participants subsequent to the removal of signals arising from ‘classical’ T2D biomarkers, notably both the α- and β-anomers of glucose, together with ketone bodies (acetone, acetoacetate and 3-D-hydroxybutyrate), the upregulated levels of these being high to very high in uncontrolled or poorly controlled T2D patients. From this analysis, using partial least squares-discriminatory analysis (PLS-DA) and further MV techniques, a Q² quality metric value of as high as 0.65 was obtained, and those attributable to differential score values on the most discriminating component (component 1) were found to be T2D-upregulated N-acetyl metabolites, creatinine/creatine and citrate, together with T2D-downregulated hippurate and indoxyl sulphate. These details are also provided in the ‘First-Steps’ review paper of Alonso-Moreno et al. [27], who devoted the whole of their Section dedicated to diabetes to a range of our group’s publications only, with no fewer than three of our Figures and one of our Tables being included therein.

Preliminary T2D monitoring experiments performed herein featured the analysis of control (non-diabetic) urine samples for the NPC1 disease investigations which were pre-spiked with increasing glucose and other metabolite concentrations for the purpose of identifying and/or confirming any present in the samples evaluated; these spectra were acquired on buffered ²H₂O-reconstituted lyophilizates of these samples. Although limited in number, with experiments being performed on small datasets only (3–4 concentration points per plot), these plots demonstrated reasonable linear relationships between TSP-normalised glucose signal intensities and added D-glucose concentration (r = 0.93–0.98). In view of their sparsity, these relationships were described as semi-quantitative plots only. Although not applicable here in view of the glucose level added, one precaution taken for future studies was the removal of samples containing glucose concentrations below the threshold range which restricts this form of analysis (≤8.0 mmol/L) [1]. Determinations at levels higher than this were, however, assisted by application of the ILSFDM deconvolution technique.

When using the necessary sample preparation technique employed for urine specimens analysed on 60 MHz benchtop spectrometers which did not offer the water signal presaturation advantage, a series of further (non-glucose) metabolites were also detectable therein, including acetate, N-acetylsugars and glycoproteins, succinate, citrate, creatinine, trimethylamine oxide, indoxyl sulphate, hippurate and formate. However, for investigations using authentic T2D urine samples performed later, with prior virtual removal of the intense water signal via lyophilization techniques, out of the T2D ketone body markers, only resonances for 3-D-hydroxybutyrate could be successfully discerned in spectra acquired. They were, nevertheless, detectable and quantifiable at an operating frequency of 400 MHz, since this approach did not require the lyophilization of samples before analysis (Figure 2) [1,4], and the sensitivity of the technique was, of course, significantly greater. However, they were also visible in 60 MHz spectra acquired on more recently developed spectrometers using water signal presaturation superiorities [1,4]. Since acetone is volatile (b.pt. 56 °C), the lyophilization step featured in early experiments is expected to remove it from the analytical platform, perhaps completely. Moreover, acetone arises from the ‘spontaneous’ decomposition of acetoacetate, this reaction also generating CO₂. Hence, this stage involved in such earlier studies limited evaluation of these two ketone bodies in T1D and T2D urine samples under these conditions. Similar results were obtained on ketone body-deplete control human urine samples which were ‘spiked’ with 50.00–200.00 mmol/L D-glucose alone. For the benefit of readers, more recent updates on the acquisition, analysis and interpretation of benchtop urinary ¹H NMR profiles from T2D patients for quantitative metabolomics analysis are available in Refs. [1,2,3,4]. Moreover, a suitable review of the applications of NMR-based metabolomics investigations to the diagnosis and tracking of human T2D in general is available in Ref. [28].

In a further diabetes-centred study, in 2020 Stolz et al. [29] investigated the capacity of LF ¹H NMR analysis to supply reliable whole blood glucose concentrations in humans. These researchers were indeed able to surpass complications arising from any superimposing signals in this biofluid, together with the near vicinity of water to specific glucose resonances. The approach selected also overcame any problems arising from the separation of cells from plasma throughout analysis durations. Therefore, from the investigation conducted, these LF NMR-based blood glucose measurements were achievable at an operating frequency of only 42.6 MHz (1.0 T), with an acceptable level of accuracy; determinations were within an acceptable ±11% range given by the German Medical Association, which was developed employing a pre-validated reference protocol.

However, nowadays there is clearly no longer much call to have such an analysis facility dedicated to determining blood glucose concentrations alone (presumably at clinical ‘point-of-care’ sites), most especially in view of the advent and prescription-ready dispensing of small wearable and easily applied Continuous or Flash glucose monitoring systems available for diabetics. Indeed, these are now extensively employed for this purpose, and hence serve to provide patients and their clinicians with invaluable data and information regarding the clinical acceptability and management of their cases [30]. Such ‘real-time’ blood glucose readings and accompanying trend datasets offer much potential regarding their benefits towards diabetes management, and hence it readily supersedes benchtop ¹H NMR analysis methods focused on blood glucose measurements alone. Obviously, the real value of compact LF NMR devices in a clinical sense is to simultaneously provide analyte datasets on whole ‘patterns’ of metabolites rather than just a single one (glucose in this case). This information supports the acquisition of valuable information on associated metabolic markers (e.g., those with the same principal component (PC) sources or metabolic pathways) or, alternatively, linked comorbidities such as chronic kidney disease (CKD) and other nephrotic disorders, and/or cardiovascular diseases for diabetic conditions.

5.2. Value of Compact NMR Analysis of Biofluids for the Diagnosis of Tuberculosis (TB)

A series of metabolomics studies performed by one group have been focused on the applications of benchtop ¹H NMR spectroscopy to define a urinary metabolic profile for human patients and animals with tuberculosis (TB) [31,32,33]. The first of these was found to offer potential for discriminating between TB and pneumococcal pneumonia, latent TB and uninfected controls, and the investigators involved discovered that it may provide a new means for the identification of non-sputum-based biomarkers for TB diagnosis [31]; the second continued these investigations in 2021, and involved an ¹H NMR-linked fingerprinting investigation for diagnosing TB in children, although it showed only very low disease confirmation rates when spectra were obtained on both LF and HF spectrometers (resonance overlap in the LF spectra acquired further limited its diagnostic potential) [32]; and thirdly, a combination of both LF and HF technologies employed in an attempt to compare the blood plasma profiles of cattle diagnosed with TB with those which were either paratuberculosis (PTB)-vaccinated, unvaccinated healthy controls or PTB-positive [33], and the researchers involved discovered that both spectrometer-based metabolomics approaches were effective in distinguishing between these animal classifications, which was considered to be a most valuable observation since both PTB-positive and PTB-vaccinated animals can give rise to a false positive TB diagnosis when tested by conventional methods.

5.3. Benchtop NMR Spectroscopic Applications to Chronic Kidney Disease Diagnosis in Cats

One further study reported by Finch et al. [34] was based on the applications of benchtop NMR spectroscopy to investigate biomolecular profiling in urine samples collected from felines with CKD. However, this investigation involved the analysis of samples collected from only two CKD and two healthy control cats, which is clearly not acceptable in view of minimal sample size requirements required for the performance of ‘state-of-the-art’ metabolomics investigations. Indeed, one general ‘rule-of-thumb’ amongst metabolomics scientists is that you require ca. 3-fold the numbers of screening samples for evaluation per group or classification for every metabolic ‘predictor’ variable considered, so for this CKD study, that yields a minimum of 3 × 17 = 51 per group [35]. However, a further general recommendation is that a minimum of 20 biofluid samples per group is required, although a generalised and rapid review of the literature has revealed that these numbers are rarely exceeded in many metabolomics experiments. Although we accept that urinary creatinine levels in CKD felines are significantly pathologically elevated, this is one of the very few conclusions that could be drawn from this sparse dataset-based experiment, especially without any recommended, spectrally verified metabolite quantification steps made therein. However, it is important to note that ²H₂O present in the analytical solution medium (20% (v/v)) can, at least in principle, diminish the intensity of creatinine’s CO-CH₂-NH proton resonance at δ = 4.05 ppm, through its exchange with solution-phase deuterium [36] (as discussed below in Section 6.3).

5.4. Benchtop NMR-Based Metabolomics Investigation of Colitis in Mice

Another research group investigated the application of benchtop NMR analysis to evaluate the detection of dextran sodium sulphate (DSS)-induced colitis in mice, using aqueous extracts of solid matrix faeces [37], and for this exercise ca. 250–300 mg of powdered faeces were weighed and then admixed with a 1:4 (w/v) ratio of 50 mmol/L phosphate-buffer solution with 0.004% (w/v) sodium azide, and 10.00% (w/v) ²H₂O containing 0.50 mmol/L TSP, plus 1.00 mmol/L sodium formate, which both served as internal standards. Notably, however, faeces do not strictly qualify as ‘biofluid media’ in the true sense of the word. Furthermore, formate may itself serve as a significant biomarker, perhaps for the activity and proliferation of selected, microbiological populations, and hence its addition as an internal standard in the described study may prove to be a significant limitation. Following shaking and centrifugation, the resulting supernatants were collected and then ultrafiltered with a 5 kDa threshold centrifugal filter. In the current paper’s authors’ view, this quantity of faeces collected, and its extraction for analysis, are clearly sufficient to obtain reliable ¹H NMR profiles with more than acceptable signal-to-noise (STN) values, even at LF operating frequencies.

In total, 19 biomolecules were assigned in the 60 MHz ¹H NMR profiles obtained, and untargeted MV analysis was able to effectively distinguish the colitis group from the healthy control one. A good level of comparability with high-field NMR analysis conducted at an operating frequency of 800 MHz was observed. Moreover, it was also found that the concentration of faecal acetate (representing a metabolite with characteristic behaviour), could be accurately quantified in 60 MHz spectra using a generalised Lorentzian curve fitting model.

Results acquired also from both the 60 and 800 MHz ¹H NMR profiles provided evidence for increased faecal concentrations of acetate and succinate in the DSS-induced mice. Such differences were also manifested as colitis-mediated decreases in the levels of n-butyrate, glycerol, alanine, aspartate, threonine and BCAAs. However, one major limitation from the spectra acquired at 60 MHz was, as expected, resonance superimposition complications from adjacent resonances, a factor which presents a barrier concerning the identification and most especially quantification of biomolecules present in such biosamples. Therefore, much attention needs to be paid to these problems, as we have previously suggested [1,2,3]. Furthermore, it is known that some metabolites have quite different resonance J-coupling splittings in low- and high-field strength spectra (e.g., propionate) [4]. Although careful spectral bucketing may serve to partially circumvent this issue, analytical data acquired do remain significantly affected.

5.5. Neonatal Sepsis in Humans

Neonatal sepsis represents an infection-mediated systemic inflammatory response syndrome which arises in both premature and term neonates [38], and serves as one of the moist prevalent causes of neonatal death and morbidity. Furthermore, it appears to play a major role in the majority of inflammatory conditions which induce or promote important morbidities which influence the pre-term, these featuring white matter injury, bronchopulmonary dysplasia, necrotizing enterocolitis, and prematurity retinopathy. Generally, newborn sepsis disorders are grouped according to their onsets, either early-onset sepsis in newborns (with infection occurring ≤3 days after birth), or late onset sepsis where it develops subsequently [38].

In 2023, Stocchero et al. [39] assessed the value of a LF compact NMR spectrometer to provide metabolic fingerprinting information for the early detection of sepsis in urine samples collected from pre-term newborn human participants, and for this purpose reference was made to a previous study conducted using untargeted mass spectrometric (MS) analysis. Interestingly, the LF NMR-based classification system produced was found to behave comparably with the MS-based system, most notably when differing biomarker classes were considered. From these investigations, and considering LF NMR spectral regions of interest, a HF NMR technique then generated a set of discriminatory metabolites which were all relevant to early onset sepsis, and this included those which although not discovered by MS analysis, they were reported as pertinent in other reported studies. These researchers also found a strong correlation between the LF and HF spectral profiles obtained, and concluded that analysis using the benchtop facility offers a propitious concept for the observation of early-onset sepsis.

5.6. Benchtop NMR Estimation of Inflammatory and Cardiovascular Disease Markers in Human Blood Plasma and Serum

In 2022, Nitschke et al. [40] performed a J-edited diffusional (JEDI) ¹H NMR pulse experiment on a LF compact 80 MHz spectrometer device which featured selective relaxational, diffusional and J-modulation signal editing for the purpose of quantitatively monitoring two newly discovered blood serum markers of SARS-CoV-2 infection and inflammation in general. These JEDI spectra collectively provided an idiosyncratic composite pattern of important biomarker resonances arising from (1) N-acetylated glycoproteins (GlycA and B) and (2) a supramolecular phospholipid composite (SPC) which are, in relative terms, relatively amplified by the above properties of the JEDI strategy applied, that also markedly diminished inputs from further plasma biomolecules. Particularly valuable for diagnostic purposes was the SPC/Glyc ratio, which was found to be virtually identical in 80 and 600 MHz spectral profiles acquired; these ratios were very highly significantly different between SARS-CoV-2 positive and healthy control patients (p < 10⁻⁷ for both spectrometers utilised). The authors of this work also support the installation of such near-portable LF devices at sites which do not require a sophisticated laboratory support set-up, and which successfully circumnavigates the hurdle to its direct clinical applications, as we have previously suggested [1,2,3,4]. Notably, the SPC resonance, which represents part of this unique inflammatory pattern, was able to furnish a direct ‘model-free’ determination of both low-density-lipoprotein (LDL) and high-density-lipoprotein (HDL) major fractions via signal integration.

This group also reported the further development and transition of this bioanalytical approach from HF to LF benchtop NMR for a comprehensive blood plasma or serum lipoprotein analysis using the latter technique, some of the analytes parameters determined serving as cardiovascular risk factors [41]. For that investigation, a quantitative calibration scheme was employed to obtain durable and reproducible results acquired at different laboratory sites, irrespective of LF spectrometer-mediated lower spectral dispersion and overall sensitivity. This admirable work demonstrated that irrespective of operating frequency and its diminished spectral dispersion and sensitivity for LF benchtop devices, a total of 25 out of 28 major lipoprotein markers could be determined, which encompassed some key cardiometabolic risk characteristics. The authors concluded that the strategy employed had potential for direct molecular phenotyping at clinical locations, and potentially also for longitudinal personalised medicine.

6. Limitations of Our Original 2014 Benchtop NMR Study, and Longitudinal Improvements in the Technology Since That Time

In view of its novel pioneering nature, the early investigation conducted in 2014 described here was not without its limitations. Perhaps its most obvious limitation was the absence of field-gradient-based solvent suppression systems, which are now more than essential when employing LF benchtop NMR technologies for metabolomics purposes. Notwithstanding, currently such advantages now almost routinely suppress interferences and overlap of the highly intense H₂O/HOD signal with many of those of interest, and therefore enhances the integrity of such investigations.

Moreover, as noted above, one further major drawback of the applications of benchtop ¹H NMR spectroscopy to the analysis of complete biofluid profiles at these early pioneering stages was the unavoidable superimposition of biomolecule resonances of interest. However, as detailed in [1,2,3,4], this effect remains today, although it should be noted that a range of attempts have been made to surpass this analytical barrier and improve quantitative NMR analysis, for example, through the application of ‘state-of-the-art’ resonance deconvolution techniques such as those regularly used for lipoprotein analysis at HF and now LF operating frequencies [40,41].

Of special relevance to this work, the reliable and robust suppression of such intense solvent (predominantly water)-based resonances has certainly captivated much interest and discussion within the NMR and NMR-linked metabolomics research communities for a lengthy period [27,42,43]. The most ideal and effective strategies for this are, of course, determined by the sample types, their pH values and the molecular masses of sample analytes considered. Critically, since all LF benchtop NMR analysis devices have an inherent marked decrease in spectral dispersion, the solvent suppression bandwidth applied must be sufficiently narrow and effectual, so that its influence on adjacent ¹H signals is diminished as much as possible.

From this early study, key points established by our group for the future consideration of biofluid metabolites for inclusion in future metabolomics studies were that resonance ‘buckets’ should be restricted to those containing only prominent, relatively intense signals with methyl or even methylene group singlets, doublets or triplets, and that each one should be present in relatively ‘free’ regions of spectra in order to avoid interferences from adjacent resonances, to a maximum levels of say 5%, but certainly no more than 10% overlap threshold (approximately corresponding to the + superimposition status provided in Table 1). However, this deduction was made prior to the consideration and evaluation of NMR deconvolution software to avoid such resonance overlap, a research area which has, to date, been very rewarding. Currently, our group utilises the ILSFDM software [8] coupled with manual bucketing, almost routinely, where required, although more powerful software and AI-based new developments are now available, as outlined in Section 6.1 and Section 6.2.

A further critical consideration is the limited sample sizes employed for these experiments, which obviously represents a major weakness of these experiments. In view of (1) the limited availability of and access to such benchtop NMR spectrometer facilities when first conducted in 2014, and (2) the essential requirement to analyse only ²H₂O-reconstituted lyophilizates of human urine samples, a sample preparation stage which took some time to complete, it was only possible to conduct experiments on a limited number of samples. However, it should be noted that the major objectives of the current paper were mainly targeted on the very first acquisition of at least reasonable quality, reliable and reproducible spectral profiles of human biofluid samples using LF benchtop NMR technologies at that point in time (late 2014), experiments potentially leading to the reliable quantitative determination of biofluid metabolites so that they could, in principle, augment the future MV benchtop NMR-based metabolomics analysis of biofluids.

In view of this study’s novelty and pioneering nature, such complications and limitations were certainly not unexpected!

6.1. Water Resonance Suppression in Biofluid Samples for Compact ¹H NMR Analysis

Since success with the use of benchtop NMR for biofluid analysis and linked metabolomics investigations is critically dependent on the incorporation of an effective solvent suppression signal for the very intense water signal present, much progress has now been made to achieve this; such advances have now been made available and implemented into standard NMR analytical operating protocols [44,45]. For our very first experiments conducted on the Oxford Instruments Pulsar device, which at that time (2014) did not offer such an advantage, it was necessary to lyophilise urine samples and then reconstitute in a buffered ²H₂O medium. Ideally, such a technique should be focused only on water (or other solvent) resonance suppression without influencing the intensities of adjacent signals, although those which do not have a pulse gradient system, notably the 1D Pre-SAT based system, can also affect neighbouring resonances, most notably the mutarotating α- and β-glucose anomeric −C1H protons, located at δ = 5.25 and 4.66 ppm, respectively (which are present at a 36:74% molar ratio in humans, respectively), and also those located nearby such as allantoin-C1H (s, δ = 5.40 ppm), uracil-C1H (d, δ = 5.79 ppm), urea-CONH₂ (broad s, δ = 5.78 ppm) and dihydroxyacetone-CH₂OH (s, δ = 4.41 ppm) to a lesser extent, for example. Indeed, the above glucose signals are markedly affected by application of the Pre-SAT scheme both at low- and high-field strengths. Indeed, in the study described in Ref [1], we optimised the H₂O/HOD presaturation frequency at δ = 4.80 ppm using the programmed 1D Pre-Sat function for this spectrometer, and found that there was a substantial 58% difference between the observed TSP-normalised α-glucose-C1H NMR resonance intensities at 60 MHz and those computed from the known total glucose and added TSP levels. Moreover, as expected, the β-glucose-C1H signal (δ = 4.66 ppm) was affected by application of the Pre-Sat sequence at δ = 4.80 ppm much more so than the α-glucose-C1H one in view of its closer locality to this irradiation frequency. However, we have previously demonstrated that the instigation of a rigorous calibration standard solution protocol satisfactorily addresses this quantification issue, as does the use of alternative resonances from the same metabolite, if indeed they are spectrally located sufficiently remotely from the presaturation frequency employed and beyond its range of suppressant activity.

For the ¹H NMR spectral profiles acquired on T2D urine samples at an operating frequency of 400 MHz, corresponding effects were observed [1]. Indeed, the ratios of intensities of glucose’s α- and β-C1H anomeric proton resonances was 57:43 (mean ± SEM percentage α-anomer C1H signal intensity 57.11 ± 2.40%), a value which reproducibly deviated substantially from the expected value, specifically 36:64, i.e., 36% α-anomer. This confirmed that the H₂O/HOD presaturation process employed and its corresponding power setting gave rise to a marked ‘dampening’ of the β-anomer’s -C1H NMR signal intensity, which arises from its very close chemical shift locality (δ = 4.66 ppm) to that of the presaturation frequency [1]. Therefore, it appeared that the default NMR power setting of 50 dB employed for H₂O/HOD solvent suppression at an operating frequency of 400 MHz gave rise to this unexpected and erroneous anomeric ratio. In view of this, we found that substituting (reducing) the ²H₂O content of the analysis medium, along with attenuation of the presaturation power level, generated a protocol which no longer exerted an influence on both these two proximal anomeric glucose-C1H proton resonance intensities in T2D human urine samples [1]. Under these conditions, no further adjustments for water signal restraint were required. Indeed, electronic integration demonstrated that these approaches yielded excellent bioanalytical results for urinary glucose determinations in T2D patients for both anomers, which were in excellent agreement with those obtained from LF 60 MHz ¹H NMR analysis using a conventional calibration standard solution strategy.

Despite this, provided that all NMR facilities employed for urinary glucose concentrations are carefully calibrated with a set of freshly prepared glucose calibration standard solutions, particularly those analysed at only 60 MHz operating frequency, then such adverse ‘presaturation overshooting’ effects are readily circumventable. It is also likely that where detectable, the glucopyranoside-C1αH doublet signal of sucrose located at 5.40 ppm is also affected by these secondary presaturation effects, but again following of the calibration methodology outlined above will permit its quantification when analysed at LF, MF or HF operating frequencies.

Nevertheless, we are aware that some more very effective water resonance suppression strategies are available, for example, those involving binomial type constructs, and it has been suggested that these may be required for application to complex biofluid samples during benchtop ¹H NMR analysis, or at least in situations where the concentration and hence resonance intensity of water is at least several orders of magnitude greater than that of individual biomolecular analytes. Generally, such systems are only employed if simpler Pre-SAT-type water vanquishing techniques, including NOESY-PR [46], Pre-SAT [43,46] or PURGE [47] techniques, are not as successful. In such cases, the JRS or W5-WATERGATE approaches may have to be applied, and for the most taxing analysis specimens, it can be coupled with Pre-SAT [48]. Actually, W5-WATERGATE serves as the one of the most powerful and efficacious water suppression techniques available, although the ‘sacrifice’ made with its use involves a relatively wide suppressed frequency region from the water signal centroid. Notwithstanding, at low benchtop spectrometer operating frequencies, the JRS and W5-WATERGATE pulse sequences do need some quite lengthy binomial delay durations, and this gives rise to total pulse trains within the 20–30 ms sweep, a period which also allows biofluid macromolecules such as proteins to relax. This can additionally represent a bonus for many NMR-based metabolomics researchers who are only focused on small biomolecules and their metabolic routes [48]. Notably, the JRS sequences available operate very successfully, whether supported by Pre-SAT systems or not (JRS8 and JRS10 are particularly valuable). Additionally, the WET pulse sequence, which utilises shaped pulses to concomitantly suppress a range of resonances, may also be employed where required.

Hence, without careful pre-operative planning, such presaturation strategies may give rise to a major loss of quantitative information regarding resonances which are located at similar frequencies to that of the solvent. Notably, the WET180 sequence has been shown to be one of the most effective for aqueous solution media [44,49]. However, in many metabolomics investigations, the application of simple presaturation techniques, in addition to the basic 1D NOESY approach, are still routine, or at least they certainly will be so with the now markedly increasing compact benchtop NMR facility usage.

6.2. Dealing with Resonance Superimposition Challenges at LF Operating Frequencies

Analysis of NMR spectra to reliably detect signals, characterise their frameworks and confirm the molecular structures of analytes responsible for them involves a crucial stage for the quantitative (or semi-quantitative) bioanalytical applications of LF spectrometers, which is commonly known as deconvolution. Of course, this problem is much more marked for LF benchtop facilities than it is for MF or HF systems in view of the relatively wide line-widths and spectral dispersion of resonances present therein, and is essential for solving such quantitative NMR problems, including structural elucidation and identity verification options available.

Therefore, as noted above, limitations arising from the superimposition/overlap of neighbouring resonances at low operating frequencies/field strengths of benchtop spectrometers give rise to some major issues. Clearly, this limitation can easily engender major problems with their electronic integration and hence quantification, and sometimes also the assignment, of biofluid metabolites present. However, in principle, progress made with increasing spectrometer operating frequencies, e.g., 80, 90 and 100 MHz for the Bruker Fourier 80, Magritek’s SpinSolv 90, and Nanalysis 100 MHz facilities, respectively, with more powerful permanent magnets, improves this difficulty somewhat (please refer to Supplementary Materials Section S2). Additionally, since 2014, some major advances in the development and instigation of overlapped resonance deconvolution software strategies for MV data analysis purposes have been made to counter the influx of prodigious NMR datasets, and this has been facilitated by the institution of algorithms, and deep learning techniques, including image recognition and labelling neural network systems, together with the application of ultrafast 2D NMR strategies. These have all demonstrated much potential for rapid spectral preprocessing, prediction and simulation, along with the deconvolution of many resonance superimposition problems [50,51]. One major advance documented in Refs. [38,39] was focused on the translation of the HF (600 MHz) ¹H NMR analysis of blood plasma/serum to an 80 MHz benchtop facility, coupled with the application of MV statistical strategies for the determination of many already higher-field established lipoprotein parameters at this low operating frequency. Indeed, in Ref. [39], blood serum or plasma lipoprotein datasets were extracted from three different cohorts with a 600 MHz spectra-based Bruker IVDr Lipoprotein Subclass Analysis (B.I.-LISA) strategy in order to construct an 80 MHz benchtop NMR model via the application of regression methods against the LF spectra for the same biosample cohort set. These 80 MHz spectral profiles were employed to construct a regression model for as many as 112 lipoprotein parameters derived from application of the B.I.LISA technique, and a regularised generalised canonical correlation analysis (RGCCA) [39], together with its sparse implementation variable selection process (SGCCA) [52], was utilised to conduct a multiblock linear regression analysis. The integrated dataset was divided into three data blocks of cardiovascular and inflammation marker interests, these being the 0.40–1.80 ppm CH block region (6109 complex points); the 1.80–2.25 ppm Glyc block region (655 complex points); and the 3.1–3.4 ppm SPC region (611 complex points). An additional fourth response data system comprised 112 lipoprotein parameters obtained from the 600 MHz ¹H NMR profiles using the B.I.LISA model.

Results from the standardisation and acquisition method, and external calibration, provided evidence that the models developed at all three cohort sites were similar. Furthermore, a combined model developed was successful in recovering 93% and 55% of main and overall lipoprotein marker parameters, respectively [39]. The authors concluded that results obtained verified a successful transition of multiparameter lipoprotein analysis from 600 to 80 MHz, and that this represented a key milestone regarding the potential placement of NMR-based diagnostic and prognostic probe systems at clinical sites of patient contact.

In addition, some researchers have applied field-invariant approaches which feed on the quantum mechanical activities of spin systems in order to improve the quantitative NMR analysis of molecules using LF benchtop NMR technologies [53,54,55].

Hence, the deconvolution of 1D LF benchtop ¹H NMR spectra represents a serious challenge for both spectrometers and their operators. In 2023, Schmid et al. [56] put forward a method for a robust, ‘expert-level quality’ deep learning-focused deconvolution algorithm for the analysis and interpretation of such experimental spectra. This algorithm arose from a neural network system which was pre-trained on synthetic spectral profiles, and customised preprocessing and pinpointing of the latter permitted reliable estimations of critical signal specifications. Intriguingly, the neural network model developed was able to effectively transfer the experimental spectra evaluated, and also had only low fitting disparities and sparse signal records for a series of complex problems, especially for crowded, high dynamic range and ‘shoulder’ signal spectral areas. It was also valuable for challenging broad resonance scenarios. The authors concluded that the algorithmic system developed represented a powerful advancement regarding the deconvolution of a range of complex spectral superimposition problems, although it was preferable to so-called ‘expert’ results.

Recently, Matvivchuck et al. [57] reported a novel model-based method designed to overcome challenges and limitations presented by the bioanalytical applications of benchtop NMR instruments. This novel strategy involved a full consideration of the quantum mechanical properties of underlying spin systems in order to define models, and then rendering this approach invariant to spectrometer operating frequency, so that this method was ideally suited to the analysis of spectral data acquired on compact benchtop spectrometers.

As noted above, of further importance, multiple overlapping signals in NMR spectra can be deconvoluted using an ILSFDM model using Lorentzian/Gaussian line shapes [8]. This allows position (d), intensity and line-width to be optimised, yielding not only the relative signal heights, but also accurate signal integrals too, and is currently frequently used by our research group. Indeed, this software option is quite tenacious, and can also be employed to fully resolve carbohydrate anomeric proton signals from that of residual water in urinary lyophilizate samples. More complex simulations can produce complete NMR spin-systems from different molecules which can be set to be present at vastly different concentrations, and then these are combined into an overlapping multi-component spectral profile. Of particular interest here is the ability to accurately simulate second-order coupling effects that produce distorted multiplets, which are indeed more pronounced in benchtop spectra acquired. In Ref. [58], our group obtained NMR spectra using Louiville calculations within NMR-SIM, a component of the Bruker Topspin software, and this accounted for full relaxation effects (T1 and T2) to produce FID data.

6.3. Potential Confounding Bioanalytical Effects of Added ²H₂O

As noted above, recently in 2022, Haslauer et al. [30] explored the effect of increasing added ²H₂O concentrations to the analytical buffer matrix of urine samples for high-resolution ¹H NMR analysis, and found that there was a major (up to 35%) reduction in the intensity of urinary creatinine’s -CO-CH₂-NH- signal (s, δ = 4.05 ppm) following 24 h. equilibration at ambient temperature when using 25% (v/v) ²H₂O. This was attributable to substitution of these ¹H nuclei by deuterium present in the ²H₂O medium employed, and therefore use of this signal to normalise all other urinary metabolite resonances, as is often the case in NMR-based metabolomics and many other chemical pathology experiments based on urine analysis, would be erroneous, since it would lead to inflated biomolecule analyte levels from the diminished ²H₂O-mediated intensities observed for this ‘normalisation standard’. In principle, creatinine’s N-CH₃ resonance (s, δ = 3.03 ppm) could be used for normalisation purposes, but this signal significantly overlaps with that of creatine, along with additional resonances from both endogenous and exogenous agents, the latter including that of the hypoglycaemic drug metformin [4].

In our original 2014 benchtop spectrometer study, which involved the reconstitution of lyophilised urine samples in a ²H₂O medium, the creatinine -CO-CH₂-NH- resonance (signal 14A) would, of course, be expected to be substantially diminished in intensity, and an examination of the intensity of this resonance when expressed relative to that of this metabolite’s N-CH₃ group (signal 10) clearly shows that it is indeed much lower than 0.67 (i.e., two-thirds its intensity), the value it should be without any ²H₂O-induced deuterium substitution.

7. Conclusions

In this decennial report, the authors provide an account of the very first experiments performed in 2014 which were successful in acquiring the very first near-reliable ¹H NMR profiles of a human biofluid on a LF (60 MHz) benchtop NMR instrument. The very first spectrum obtained then is shown (Figure 1), and this was for a urine sample collected from a patient with an LSD (NPC1 disease). Indeed, it was found to contain approximately 30 resonances for metabolite matching, and such assignments are provided in Table 1, although some of these remain tentative in view of known or potential signal overlap issues. In addition to the provision of valuable NPC1 disease-specific information (especially upregulated 3-AIB), this spectrum revealed a gastric permeability complication in this patient in view of the detection of quite high concentrations of urinary sucrose (the identity of this sugar was subsequently confirmed by MF NMR analysis).

Moreover, these prima facie sessions conducted in the current study were subsequently found to be translatable to the potential detection and identification of urinary biomarkers for a wide range of other human diseases, for example, the ready pinpointing of glucose and correlated ketone body resonance markers in urine collected from patients with diabetes conditions through experiments which were initially conducted with ‘surrogate’ diabetic samples, i.e., human controls ‘spiked’ with realistic levels of glucose to mimic uncontrolled T1D or T2D, as described in Section 2 and Section 5.1.

The current paper is then supported by updates that have been made since that time in the field of NMR-based metabolomics using LF benchtop NMR facilities as bioanalytical platforms. These have featured (1) the detection of TB in humans and animals, (2) the identification of colitis and chronic kidney diseases in animals, and (3) the rigorous determination of a large number of inflammatory and cardiovascular disease parameters in humans, (4) neonatal sepsis, in addition to those of (5) human T2D and (6) human NPC1 disease in this work. We have also discussed limitations and drawbacks of this original 2014 study, along with those associated with more recent investigations, including water signal suppression in biofluids, the loss of some important volatile analytes with the application of lyophilization before analysis (essential in 2014), the overcoming of inherent barriers presented by ¹H NMR signal overlap problems, and a creatinine deuterium-exchange complication arising from ²H₂O present in or added to the biofluid analysis medium as an analytical aid.

Critically, in order for LF benchtop NMR analysis to serve as an effective metabolomics discovery probe, it requires an effective and reliable solvent (water) resonance vanquishing scheme which performs this process with only minimal levels of attenuation of any neighbouring signals, since this may result in the loss of valuable metabolic data (the estimated water concentration of such samples is ca. 50 mol/L). With the use of such a scheme, which is certainly essential for the majority of aqueous-based metabolomics studies conducted today, resonances adjacent to the intense water signal are retained, and this allows their assignment and quantification. Fortunately, the Pre-SAT water signal suppression strategy, along with some alternative pulse sequences, have been employed for this purpose since the time of this prime study conducted in 2014. However, a series of new or further developed pulse sequences are now available to circumvent some of the barriers associated with the Pre-SAT sequence, and these offer some additional benefits, as reviewed in Section 6.2 of the current paper. Future investigations for the analysis of low-molecular-mass metabolites in protein-rich biofluids such as blood plasma may also require pulse sequences for the simultaneous removal of a broad ‘spectrum’ of rapidly relaxing biomacromolecule resonances. The benchtop NMR instrument originally employed for these studies utilised standard 5 mm diameter tubes, and hence sample handling was already familiar and quite convenient. In addition to metabolomics/biofluid analysis investigations, it is known that such spectrometers and more recently developed instruments may be deployed in common university or commercial laboratories tasked for most forms of chemical/biochemical research programmes, such as basic analytical and synthetic chemistries. Since there is no superconducting magnet featured, this novel facility costs only a fraction of funding required for more conventional medium- (300–400 MHz) and high-field (≥600 MHz) spectrometers to purchase, and obviously requires no expensive ongoing cryogen refills. Moreover, only a standard mains power supply is required, and it consumes no greater power than a regular PC facility. Since the stray magnetic field is localised entirely within the rare earth magnetic enclosure, it is rendered safe to deploy in a variety of environments, laboratory or otherwise. As demonstrated in previous studies, the compact and portable whole spectrometer dimensional advantages, coupled with the successful performance of this bioanalytical approach, renders LF benchtop NMR-based metabolic fingerprinting to be a very promising diagnostic tool. Further applications of this novel technique include those based in relatively advanced levels of chemical reaction progress monitoring, undergraduate teaching, pharmaceuticals, quality control, and environmental and forensic sciences [2].

The prima facie observation provided in this work provides a highly valuable example of the metabolomics potential of the early LF spectrometer evaluated, and prospects for the future development of such compact devices were at that time, and still are, manifold. Current expectations feature future clinically based research projects involving NMR and deep-learning artificial intelligence (AI)-linked metabolomics investigations using now much-updated 2024/2025 versions of this and other non-stationary benchtop spectrometers. Such AI-facilitated approaches will clearly serve to allow users to circumnavigate and hopefully overcome at least some of the resonance superimposition problems commonly encountered with this technique.