*2.4. Gadolinium Improves the Lower Limit of Detection and Quantification*

We next sought to determine the lower limit of detection (LOD) and lower limit of quantification (LOQ) for our NUS measurements in the presence of Gd. LOD and LOQ are defined as follows:

$$\text{LOD} = \text{3} \times \text{\textbullet} \tag{1}$$

$$\text{LOQ} = 10 \times \sigma \tag{2}$$

where the variance of the noise (σ) was estimated by the median absolute deviation (*MAD*). *MAD* was calculated from the COLMAR database [35], where the positive values for all non-peak data (*Xi*) were used in the following equations:

$$MAD = median\_i(\|X\_i - median\_i(X\_j)\|)\tag{3}$$

$$
\sigma = 1.4826 \times \text{MAD} \tag{4}
$$

Tables 2 and 3 list the LOD and LOQ for each of the resonances detected in Reference 2. Metabolites with multiple resonances have an LOD/LOQ for each observed peak, and thus metabolites with multiple peaks will have a range of LOD/LOQ values. The average LOD and LOQ in the presence of Gd was 7.8 ± 0.3 μM and 26 ± 1 μM, respectively. This is a dramatic improvement over our previous findings that yielded an average LOD and LOQ of 19.1 μM and 65.6 μM, respectively [15]. These prior NMR experiments lacked the addition of Gd and used a longer d1 of 1.5 s. Thus, it is possible to detect lower abundant metabolites by adding Gd and decreasing d1. We also compared the effects of different NMR probes on LOD/LOQ. For the same d1 of 1.5 s, a TCI helium-cooled probe had a lower LOD/LOQ compared to a TXI nitrogen-cooled probe (Table S3).

#### *2.5. Gadolinium Maintains Reproducibility*

Highly reproducible measurements are required to detect changes in the large number of samples associated with metabolomics studies. We previously demonstrated that intensity measurements from NUS 1H-13C HSQC experiments with a d1 of 1.5 s were highly reproducible as evident by a percent coefficient of variation (%CV) of 14 ± 9% for a model mixture containing 15 metabolites at a concentration of 500 μM [15]. By decreasing the d1 to 0.8 s, we observed a decrease in the %CV to 8 ± 8% (Figure 4) for three replicates of Reference 1. This was expected, given that the increased number of scans would lead to an increase in peak intensities. We only observed a modest decrease in %CV to 7 ± 7% (Figure 4) by adding Gd to the samples while maintaining a d1 of 0.8 s. This suggests that the addition of Gd does not negatively impact the reproducibility of NUS 1H-13C HSQC experiments and may increase the reliability of these measurements.
