**3. Discussion**

Metabolites are influenced both by the genome and the environment, and thus provide the most comprehensive readout for the state of an individual [36–38]. By monitoring changes in metabolites, it is possible to develop novel biomarkers that reveal important health information. Indeed, altered metabolite levels have been observed in many diseases, including diabetes [39], neurodegeneration [40], cancer [4], cardiovascular disease [6], and even aging [3]. Furthermore, in a series of separate studies, we have identified metabolite biomarkers of response (BoRs) that correlate with drug responsiveness for metastatic breast cancer patients treated with CDK4/6 inhibitors as well as the anti-HER2 therapy trastuzumab; and for gastrointestinal stromal tumor (GIST) patients treated with tyrosine kinase inhibitors [41–43]. While additional validation studies are required, these preliminary results suggest the exciting possibility that metabolite-based biomarkers have for designing optimal treatment strategies for individual patients, which is a major goal of precision medicine.

To uncover metabolite BoRs, it is first necessary to accurately measure metabolite levels in biospecimens collected from a large number of patients so that the relative metabolite concentration can be correlated with disease outcomes and/or a drug response. NMR and mass spectrometry (MS) are the two most commonly used analytical platforms for measuring metabolites. Traditionally, MS has been favored due to its high sensitivity, dynamic range, and potential for high throughput. There are numerous sensitive LC-MS methods reported in the literature for the identification of endogenous metabolites in human plasma [19–21,44–46]. For example, amino acids are routinely detected at submicromolar concentrations (0.01 to 0.04 μM) by these targeted LC-MS methods. In contrast, NMRbased approaches typically detect plasma concentrations in the micromolar (3–10 μM) range [44–46]. However, MS can suffer from reproducibility issues, requires chromatography because of the narrow molecular-weight distribution of metabolites, and still faces challenges in metabolite identification [9]. Conversely, NMR is highly reproducible and can reveal structural information to facilitate metabolite identification. However, NMR is limited by sensitivity and spectral overlap [47]. Multidimensional NMR can overcome some of these challenges but requires extremely long experimental times that are not practical for the large number of samples needed for BoR discovery. Efforts to increase the throughput of NMR are actively being explored. We and others have demonstrated that NUS can accelerate NMR acquisition times to meet the high-throughput demands of metabolomics [13–15]. In an approximate one-hour experiment, we verified that intensity measurements from NUS 1H-13C HSQC spectra are highly reproducible and can facilitate the detection of a wide variety of metabolites in the low micromolar range.

In this study, we sought to explore additional techniques to extend the limit of metabolite detection by multidimensional NMR. As a first step, we assessed the effect of the d1 relaxation delay on S/N. The relaxation delay is the experimental time between scans in an NMR experiment to allow the nuclear spins to return to equilibrium, which is influenced by T1 longitudinal relaxation rates of each nuclei in the sample. For quantitative NMR, it is suggested to set d1 to at least 5 times the slowest T1 [17]. For metabolomics, this is not practical as T1s can be several seconds in length or longer. Instead, d1 is commonly set to a shorter, predetermined value for semi-quantification. Herein, we demonstrated that a decrease in d1 from 1.5 s to 0.8 s enabled an increase in the number of scans from 36 to 72, which led to an overall improvement in S/N and an increase in the mean signal intensity for metabolite resonances. Notably, this was accomplished without increasing the total time to acquire the NMR spectrum. Any further reduction in d1 was observed to result in severe signal artifacts from the solvent.

With the optimal d1 selected, we next sought to manipulate the T1s of metabolite nuclei through PRE. PRE accelerates spin relaxation due to induced magnetic dipolar interactions with unpaired electrons. PRE-based applications have been used for macromolecular structure determination, characterizing long-range interactions and identifying transiently populated states of proteins and complexes [28,48,49]. PRE agents also provide the foundation for contrast agents in magnetic resonance imaging (MRI) [50]. Previous metabolomics studies have suggested that the addition of PRE-agents, Cu(EDTA) or Gd, can decrease T1 relaxation times for metabolites [13,17]. We observed similar trends using our standard 25% NUS 1H-13C HSQC experimental parameters and noted that the addition of Gd led to an overall improvement in S/N and mean signal intensity for metabolites in a model mixture and from plasma samples. Although the average increased fold change in intensity with Gd was relatively modest, we did observe a significant improvement (>2-fold increase) for metabolite resonances with the lowest signal intensity. For metabolomics studies, the ability to accurately detect and quantify a broad range of metabolites that span different chemical classes and concentration ranges is paramount. Thus, an increase in the NMR signal intensity for low abundant metabolites suggests that the addition of Gd could improve the coverage of the metabolome. Indeed, both the lower limit of detection and quantification (LoD/LoQ) were significantly improved in the presence of Gd. In our previous results, the LoD and LoQ for a model mixture of metabolites was 19.1 μM and 65.6 μM, respectively. For the same model mixture, the LoD and LoQ decreased by more than 2-fold to 7.8 μM and 26 μM, respectively, by decreasing the d1 and by the addition of Gd.

## **4. Materials and Methods**

Commercially available analytical standards were used to prepare model mixtures of metabolites, Reference 1 and Reference 2 (Tables S1 and S2): acetylcholine chloride (C7H15NO2·HCl, >99%), L-arginine (C6H14N4O2, >98%), L-glutamine (C5H10N2O3, >99%), D-alpha-hydroxyglutaric acid disodium salt (C5H6Na2O5, >98%), α-ketoglutaric acid disodium salt dihydrate (C5H4Na2O5·2H2O, >98%), adenosine 5-monophosphate disodium (C10H12N5Na2O7P, >99%), D-(-)-fructose (C6H12O6, >99%), guanosine 5-triphosphate sodium salt (C10H16N5O14P3·xNa + yH2O, >95%), lithium potassium acetyl phosphate (C2H3KLiO5P, >97%), L-ornithine hydrochloride (C5H12N2O2·HCl, >98%), β-nicotinamide adenine dinucleotide hydrate (C21H27N7O14P2·xH2O, >98%), DL-malic acid (C4H6O5, >99%), D-ribose 5-phosphate disodium salt dihydrate (C5H9Na2O8P·2H2O, >99%), sodium succinate dibasic hexahydrate (C4H4Na2O4·6H2O, >99%), sodium acetate (C2H3NaO2, >99%), sodium L-lactate (C3H5NaO3, >99), sodium citrate tribasic dihydrate (C6H5O7Na3·2H2O, >99%), sodium fumarate dibasic (C4H2Na2O4, >98%), sodium pyruvate (C3H3NaO3, >99%), uridine 5-diphosphate (C9H12N2Na2O12P2·xH2O, >96%), Lalanine (C3H7NO2, >98%), L-cysteine (C3H7NO2S, >98%), D-(+)-glucosamine hydrochloride (C6H13NO5·HCl, >99%), D-(+)-glucose (C6H12O6, >99.5%), choline chloride (C5H13NO·HCl, >99%), cytidine (C9H13N3O5, >99%), L-leucine (C6H13NO2, >98.5%), L-glutamic acid monosodium salt monohydrate (C5H8NNaO4·H2O, >99%), L-histidine (C6H11N3O3·HCl, >98.5%), L-lysine, monohydrochloride (C6H14N2O2·HCl, >98.5%). All the compounds were obtained from Sigma-Aldrich. Deuterium oxide (D2O, 99.0%) was purchased from Cambridge Isotope Laboratory, Inc., Andover, MA. Pooled human plasma (apheresisderived, K2EDTA) was purchased from innovative research, Novi, MI. Paramagnetic relaxation agents gadobutrol (Gd) (C18H31GdN4O9, >99.9%) and copper (II) disodium ethylenediaminetetraacetate tetrahydrate (Cu- EDTA) (C10H12CuN2Na2O8·4H2O) were procured from MedChemExpress, Monmouth Junction, NJ and TCI America, Portland, OR respectively. The NMR reference standard, deuterated 3-(trimethylsilyl)-1-propanesulfonic acid sodium salt (DSS-d6, 98%) was purchased from Cambridge Isotope Laboratory, Andover, MA, USA.
