Integrating NMR and multi-LC-MS-based untargeted metabolomics for comprehensive analysis of blood serum samples

Liquid chromatography coupled to mass spectrometry (LC-MS) has emerged as the predominant analytical technique used in metabolomics [[1], [2], [3],[11], [12], [13]], with reversed-phase LC-MS (RPLC-MS) [14,15] being the most widely applied approach, albeit with limitations for the analysis of highly polar metabolites, which are often the most abundant in cells and biofluids [16,17]. Alternative LC-MS retention mechanisms are often used to characterise these metabolites, including mixed-mode chromatography-MS [18], hydrophilic interaction chromatography-MS (HILIC-MS) [19,20], ion-exchange chromatography-MS (IC-MS) [16,17,21,22], and combinations of these methods to obtain broader metabolome coverage. While LC-MS-based metabolomics offers high compound coverage with low sample volume requirements, challenges such as susceptibility to batch effects, low compound-feature number to metabolite identification ratio, and selective quantification capabilities persist [1,23]. On the other hand, despite its lower sensitivity, nuclear magnetic resonance (NMR) provides highly reproducible and quantitative analysis, is non-destructive and can measure peptides, lipoproteins, glycoproteins and other small proteins, in addition to low molecular weight metabolites [[24], [25], [26], [27], [28], [29]]. Combining NMR and LC-MS approaches, therefore, has the potential to significantly enrich metabolite coverage [[30], [31], [32], [33], [34], [35]]. The Venn diagram in Fig. 1 represents indicative coverage of different compound classes reported using the complementary analytical platforms: NMR, RPLC-MS, and IC-MS [16,27,29,30,[36], [37], [38], [39], [40]].

Analytica Chimica Acta, Volume 1356, 2025, 343979: Fig. 1. The scope and overlap of the individual analytical platforms addressed in this study. (HDLs – high-density lipoproteins; LDLs – low-density lipoproteins; PUFAs – polyunsaturated fatty acids; BCAAs – branched-chain amino acids; SCFAs – short-chain fatty acids; AAs – amino acids).

Given the potential benefits of combining NMR with comprehensive multi-LC-MS analysis for enhanced metabolome coverage and biomarker discovery in a research context, we explored ways to prepare serum samples that would allow for sequential NMR and MS analysis. Our primary aim was to develop a sample preparation method for untargeted metabolomics using a single sample aliquot suitable for analysis on both NMR and multiple LC-MS platforms. To do so, we explored sample preparation using deuterated solvents and sought evidence of whether this led to deuteration across the serum metabolome. Furthermore, we investigated changes in the LC-MS metabolome in response to the addition of NMR buffer to serum samples. Finally, we compared different approaches for protein removal (precipitation in organic solvent and MWCO filtration) and their impact on serum metabolome coverage. This led to an optimised blood serum sample preparation workflow for untargeted metabolomics suitable for a research context, prioritising broad metabolite coverage and use of minimal sample volumes. This approach facilitates metabolite discovery and highlights key considerations for implementing a multiplatform strategy.

2. Materials and methods

2.6. Anion-exchange chromatography-MS analysis

Anion-exchange chromatography-MS analysis was carried out using a Dionex ICS-5000+ high-pressure ion chromatography system equipped with a continuously regenerated trap column, Dionex ERS 500e suppressor and AS11-HC (2 × 250 mm, 4 μm) column, from Dionex (Sunnyvale, CA, USA), coupled to a Q-Exactive hybrid quadrupole-Orbitrap MS via a HESI II probe (Thermo Fisher, San Jose, CA, USA). The technical parameters for the untargeted metabolomics method have been previously published in Ref. [17].

2.7. Reversed-phase ultra-performance liquid chromatography-MS analysis

C18 reversed-phase analysis was performed using an Acquity UPLC liquid chromatograph (Waters, UK) system with a gradient elution program coupled directly to a high-resolution tandem mass spectrometer, Xevo G2-XS QTOF (Waters, UK). A 5 μL partial loop injection was used for all analyses with pre- and post-injection wash programs. A Waters CORTECS UPLC T3 column (2.1 × 100 mm, 1.6 μm; Waters PLC, Milford, MA, USA) was used with a flow rate of 0.3 mL/min. The total run time was 18 min. Mobile phase A comprised Milli-Q water with 0.1 % formic acid, and mobile phase B was 100 % methanol with 0.1 % formic acid. The gradient elution program was as follows: 0 min, 5 % B; 4 min, 50 % B; 12 min, 99 % B; 15 min, 99 % B; 15.1 min, 5 % B; 18 min, 5 % B. The column temperature was kept at 40 °C throughout the experiment. Mass spectrometry analysis was performed separately in positive and negative ion modes using a scan range from m/z 50–900. The centroid data were collected in MS^E mode, and the time was set to 0.2 s. All the parent ions were fragmented with an energy ramp of 20–40 eV, and the fragment information was collected. Tune file source parameters were set as follows: source temperature 140 °C; cone gas flow 50 L/h; desolvation gas flow 800 L/h; cone voltage and capillary voltage were set to 40.0 V and 2.5 kV in positive (pos.) ionisation mode and 40.0 V and 2.0 kV in negative (neg.) ionisation mode, respectively. Full scan data were acquired in continuum mode.

2.8. LC-MS data pre-processing and analysis

Unprocessed chromatographic and MS data were examined using vendor-specific LC-MS software - MassLynx 4.2 (Waters, Elstree, UK) for RPLC-MS and FreeStyle 1.8 (Thermo Fisher Scientific, Waltham, MA, USA) for IC-MS. The untargeted metabolomics data were then processed in Progenesis QI (Nonlinear Dynamics, Waters, Elstree, UK), and the detected compound-features were compared between the different extraction methods. A % coefficient of variation (CV) cut-off of 30 % was used for routine filtering of the untargeted metabolomics data and for evaluating the efficacy of sample preparation efficiency across independent datasets.

To compare the abundance and variability of metabolites, compound-features were matched by both retention time and m/z across the different sample preparation methods. Metabolite identification was based on matching multiple experimental data to metabolite databases. In-house libraries of over 450 (IC-MS) and 150 (RPLC-MS) authentic metabolite standards with matching of multiple independent measurements under the same experimental conditions to provide Level 1 identifications (e.g. retention time (R_t error <2 min; accurate-mass <5 ppm (IC-MS) and 20 ppm (RPLC-MS)); isotope and fragmentation patterns >90 %).

3. Results and discussion

3.2. NMR phosphate buffer induces some changes in compound-feature abundances by LC-MS

In NMR metabolomics studies, deuterated aqueous buffers, such as phosphate buffer, are commonly used to ensure high chemical shift reproducibility between samples, mitigating the influence of even slight variations in pH (note: the buffer pH is usually kept at serum levels of around pH 7.4). Conversely, MS sample preparation, typically avoids the addition of buffers due to their potential to induce chemical artifacts and/or ion suppression, thereby affecting sensitivity and peak area reproducibility. It is worth noting that there is an exception with anion-exchange chromatography-MS, where hydroxide ions are employed for gradient elution and subsequently removed through electrochemical suppression prior to MS analysis [16]. Very few studies have investigated the impact of NMR buffers on compound-feature profiles using untargeted LC-MS methods. Our objective was, therefore, to assess the effect of phosphate buffer in samples on analyte chromatographic peak area reproducibility across the metabolome using both IC-MS and RPLC-MS platforms.

3.3.1. Protein removal improves the detection of selected signals but also the loss of several serum metabolites

To evaluate the two strategies (i and ii), we assessed the impact of protein removal methods on the NMR spectral profile. Initially, we compared metabolite extraction from serum in NMR phosphate buffer with and without 10 kDa MWCO filtration. Total protein content (assessed by UV/vis spectrometry) showed a reduction from 1750 ± 80 μg/mL to 130 ± 10 μg/mL after filtration. Subsequent NMR analysis revealed a significantly improved spectral baseline by eliminating the residual signals arising from albumin, glycoproteins, and lipoproteins, particularly in the 0.8–2.0 ppm region (Fig. 7A). The removal of undesired signals, including the propylene glycol contaminant originating from blood collection tubes, was also achieved. Importantly, 10 kDa MWCO filtration facilitated the detection of specific metabolites, including branched-chain amino acids (BCAAs) and lactate, previously obscured by broad lipoprotein signals in unfiltered samples (Fig. 7B). Additionally, signals such as 2-hydroxybutyrate (0.90 ppm) and lysine (1.38 ppm, 1.73 ppm) were revealed only after the filtration step. However, MWCO filtration also led to a reduction in additional NMR spectral features, including lipoproteins, N-acetyl glycoproteins and medium-chain fatty acids (2.02 ppm). Furthermore, in filtered samples, the spectral region from 3.0 to 4.0 ppm showed chemical shifts associated with glycerol (3.54–3.58, 3.64–3.67, 3.76–3.80 ppm), released from the MWCO filter coating during filtration (Fig. 7C–SI Figs. 9–14). The glycerol multiplets interfered with carbohydrate signals and entirely overlapped with the glycine signal (3.56 ppm). Attempts to mitigate this interference through pre-washing MWCO filters using Milli-Q water and various organic solvents showed partial reduction but did not entirely eliminate the glycerol signal from the resulting spectra (SI Table 3; SI Figs. 9–12). Thus, while filtration effectively removes significant overlap in the NMR spectrum, enabling the measurement of additional low molecular weight compounds, it also results in the loss or obscuring of several biologically meaningful metabolites.

Analytica Chimica Acta, Volume 1356, 2025, 343979: Fig. 7. 1H CPMG NMR spectra of serum samples illustrating the effect of MWCO filtration. (A, B) The 0.7–2.2 ppm spectral region of (A) a 10 kDa MWCO-filtered sample and (B) an unfiltered serum sample in aqueous phosphate buffer. Panel A, compared to panel B, highlights the absence of lipoprotein, PUFA (δ 0.86, 1.28, and 1.99 ppm), N-acetyl glycoprotein (GlycA/B, δ 2.00–2.10 ppm), and propylene glycol contaminant (released from blood collection tubes, δ 1.17 ppm) signals following filtration. (C, D) The 3.1–4.2 ppm spectral region of (C) a 10 kDa MWCO-filtered sample in acetonitrile (distorted by glycerol contamination peaks at δ 3.56 (m), 3.65 (m), and 3.78 (m)); and (D) an acetonitrile-precipitated sample, which contains only physiological glycerol signals. Similar to MWCO-filtered samples in aqueous phosphate buffer, the acetonitrile-precipitated samples also lack signals corresponding to lipoproteins, PUFAs, and N-acetyl glycoproteins. For complete NMR spectra of all extraction methods, refer to SI Figs. 13 and 14.

As an alternative method for the removal of protein from serum samples, the addition of acetonitrile to precipitate protein (serum: acetonitrile ratio of 1 : 2.33 (v/v)) was also tested (Fig. 7D). Although less efficient in total protein removal (levels of 210 ± 20 μg/mL were measured after precipitation compared to 130 ± 10 μg/mL for MWCO), the NMR spectra closely resembled those of MWCO-filtered samples (SI Fig. 13B vs 14B). Both protein removal methods effectively eliminated macromolecular signals, including lipoproteins, N-acetyl glycoproteins and medium-chain fatty acids. The implementation of MWCO filters further resulted in a decreased relative abundance of lysine (1.40 ppm; p = 0.008), acetate (1.92 ppm; p = 0.003), glutamine (2.35 ppm, p = 0.02), and citrate (2.52–2.56; 2.64–2.68 ppm, p = 0.01 (note that the citrate signal was enhanced by the presence of sodium citrate-based anticoagulant)).

In conclusion, our findings related to protein removal indicate that MWCO filtration and protein precipitation using acetonitrile reduce the protein content in serum samples by approximately 95 % and 90 %, respectively. In both scenarios, there was a complete elimination of signals corresponding to lipoproteins, N-acetyl glycoproteins, and medium-chain fatty acids observed in the NMR spectra (Fig. 7A). Furthermore, the analysis of CPMG spectra for individual extractions revealed a reduction in the relative abundance of certain metabolites associated with MWCO filtration of serum samples. Conversely, protein removal enabled the detection of certain compounds that were obscured in unfiltered samples. Our observations thus highlight both advantages and disadvantages associated with protein removal prior to NMR analysis. The incorporation of the protein removal step may prove beneficial for studies exclusively focusing on low-molecular-weight metabolites, as it enhances compound detection by eliminating broad overlapping signals. Conversely, for clinical studies, quantifying macromolecular signals, such as lipoproteins, N-acetyl glycoproteins, and medium-chain fatty acids, may be desirable, given that they could serve as useful diagnostic biomarkers. Indeed, this has been demonstrated in various studies of conditions, including cancer, autoimmune disorders, and cardiovascular and neurological diseases [38,55,[71], [72], [73]].

4. Summary and conclusions

Our experiments have demonstrated that serum samples prepared for NMR metabolomics can be used for subsequent untargeted LC-MS metabolomic analysis. We have established that a single blood serum sample aliquot can be sequentially prepared for both NMR and multi-platform LC-MS-based untargeted metabolomics. Furthermore, our observations indicate minimal variations in the abundance of metabolites identified from our in-house database, thereby confirming the robustness of our preparation protocol for comprehensive metabolomic profiling. After protein removal, an NMR sample can be directly utilised for LC-MS analysis with multiple chromatographic methods, without the need to replace solvents, making this approach highly efficient and well-suited for research settings. Deuterated solvents and buffering salts required for NMR were well-tolerated in both RPLC-MS and IC-MS, with no evidence of metabolite deuteration, as demonstrated by the direct comparison of extractions using Milli-Q and deuterated water. While phosphate buffers have shown no adverse effects over two years of use, regular maintenance is recommended to mitigate the risk of salt accumulation in high-throughput workflows. We found that protein removal by acetonitrile precipitation and MWCO filtration significantly altered the NMR metabolome profile (e.g. by removing lipoprotein signals but also revealing some low-abundance molecule chemical shifts). We also demonstrated that there were differences in the metabolite profile obtained from acetonitrile-precipitated samples and those filtered by MWCO filtration, with the latter resulting in the loss of more lipophilic compounds.

We thus recommend two alternative workflows with the optimal approach dependent on research objectives. A notable advantage of these integrated NMR and LC-MS workflows lies in their efficiency, achieved through the utilisation of a single sample aliquot for all analytical platforms. This approach is particularly advantageous in research settings where retrospectively collected samples stored in biobanks with limited volumes are often used, and broad metabolite coverage is essential for metabolite discovery. Additionally, this method enables a more robust integration of NMR and MS results in untargeted metabolomics studies, enhancing metabolite coverage and providing internal validation of metabolite characterisations where overlap occurs. In conclusion, the described workflows support the routine integration of NMR and LC-MS analysis in metabolomics, facilitating efficient and comprehensive metabolite profiling.