Robust detection of oncometabolic aberrations by 1H-13C heteronuclear single quantum correlation in live cells and intact tumors ex-vivo

Extensive efforts have been made to use non-invasive 1H magnetic resonance (MR) spectroscopy to quantify metabolites that are diagnostic of specific disease states. Within the realm of precision oncology, these efforts have largely centered on quantifying 2-hydroxyglutarate (2-HG) in tumors harboring isocitrate dehydrogenase 1 (IDH1) mutations. As many metabolites have similar chemical shifts, the resulting 1H spectra of intact biological material are highly convoluted, limiting the application of 1H MR to high abundance metabolites. Hydrogen-Carbon Heteronuclear single quantum correlation 1H-13C HSQC is routinely employed in organic synthesis to resolve complex spectra but has received limited attention for biological studies. Here, we show that 1H-13C HSQC offers a dramatic improvement in sensitivity compared to one-dimensional (1D) 13C NMR and dramatic signal deconvolution compared to 1D 1H spectra in an intact biological setting. Using a standard NMR spectroscope without specialized signal enhancements features such as magic angle spinning, metabolite extractions or 13C-isotopic enrichment, we obtain well-resolved 2D 1H-13C HSQC spectra in live cancer cells, in ex-vivo freshly dissected xenografted tumors and resected primary tumors. We demonstrate that this method can readily identify tumors with specific genetic-driven oncometabolite alterations such as IDH mutations with elevation of 2-HG as well as PGD-homozygously deleted tumors with elevation of gluconate. These data support the potential of 1H-13C HSQC as a non-invasive diagnostic tool for metabolic precision oncology.


Introduction
Metabolic vulnerabilities are emerging as viable therapeutic targets within the framework of precision oncology [1][2][3] . Well-studied examples of such aberrations include tumors that have mutations in isocitrate dehydrogenase 1 and 2 (IDH1/2) enzymes. As key drivers of tumorigenesis in diverse cancers such as glioma, cholangiocarcinoma, and leukemia, Here, we demonstrate that it is possible to overcome the specificity-limitations of 1 H MRS through 1 H-13 C HSQC 21 . This technique de-convolutes the spectrum into 2 dimensions and harnesses the differences in both 1 H and 13 C chemical shifts to resolve the signalpeaks of closely related chemical species. Rather than merely peaks in a 1D spectrum, the identity of specific metabolites is characterized by a pattern or constellation of C-H HSQC peaks in a 2D area. The sensitivity of this technique is dramatically higher than standard 13 C-spectroscopy, as the signal is ultimately generated through 1 Hmagnetization as well as ultimately read in the 1 H channel 22 . We demonstrated that a phase-sensitive HSQC pulse sequence with short scan times enables detection of 2-HG and 6-phosphogluconate/gluconate in live cancer cells as well as tumors ex-vivo; this way, we can reliably identify tumors/cell lines that carry IDH-point mutations or PGDhomozygous deletions. This was achieved without signal enhancing devices (e.g., spinning), 13 C-isotope enrichment, or biochemical metabolite extraction -all despite the high magnetic in-homogeneities inherent in live cells or tumor chunks. These data suggest immediate utility in the setting of just-in-time diagnosis during tumor resection as well as the potential for use in non-invasive MRS setting in vivo.

Convolution of 1 H spectra hinders robust detection of oncometabolites in intact biological samples.
To provide a yardstick for the usefulness of 1 H-13 C HSQC in the intact biological setting, we first performed detailed studies with conventional 1D 1 H nuclear magnetic resonance (NMR). We performed proof-of-principal experiments in a standard NMR spectrometer (see methods) with standard 5 mm NMR tubes. To provide a solid point of reference, analysis of the 1 H spectrum of the 2-HG pure chemical standard in 10% D2O phosphate-buffered saline (PBS) was conducted under the exact experimental conditions subsequently used for biological experiments. The 1 H spectrum of the 2-HG standard showed a quartet at 3.92-3.95 ppm for the H2, a convoluted multiplet at 2.11-2.23 ppm for H4 and H4', and multiplets at 1.71-1.79 ppm and 1.87-1.95 ppm for H3 and H3' (Supplementary Fig. S1C). We then took 1 H scans of live mutant IDH1 cancer cells (HT-1080, NHA mIDH1, SNU-1079 and COR-L105; for a full description of the cell types and mutations, see methods). As a negative control, we also took 1 H NMR scans of IDH wildtype cell lines treated with the IDH1 inhibitor AGI-5198 23 , which inhibits the production of 2-HG. We found that 2-HG peaks are indistinct, due to broadening and overlap with peaks from other highly abundant metabolites in normal brain and wildtype IDH tumors (Supplementary Fig. S1d-S1j). For example, glutamine, glutamate, and GABA have chemical shifts close to H4 and H4' of 2-HG, while NAA resonances close to frequencies of H3 and H3' protons of 2-HG. Lactate, myo-inositol and phosphocholine and choline have chemical shifts close to H2 peak of 2-HG 19 . Though there are vague hints that peaks may be attributed to 2-HG upon post-hoc analysis, their small signal-tonoise ratio (SNR), broadness, and low amplitude compared to neighboring peaks, in no way, enable confident discrimination between IDH-mutant and WT tumors without previous genetic assignment. Furthermore, the peak broadening is likely to be worse in the MRS setting in vivo, making the confidence in these peaks for specific identification of 2-HG/IDH mutant tumors even less appealing.

H-13 C HSQC allows reliable detection of 2-HG and typing of IDH mutant status in
intact biological samples even with natural abundance 13 C. To resolve the issue of spectral overlap, we reasoned that metabolites with similar 1 H NMR shifts may be discriminated by the differences in their 13 C NMR shifts via 2D 1 H-13 C HSQC. After screening various pulse sequences available in Bruker TopSpin (3.5) MD Anderson's NMR core, we found that the phase-sensitive HSQCETGPSISP3.2 pulse program [24][25][26] yields high SNR, sharp peaks in the 13 C axis, and distinguishes between positive phase (CH3 and CH peaks) from negative phase (CH2 peaks) -shown as different colors (Fig.   1a). Together, these qualities provide an additional layer of confidence in correct metabolite-peak assignments. Using this pulse program, we first characterized the 1 H-13 C HSQC spectrum of the 2-HG chemical standard under the same conditions that we would apply to live cells (Fig. 1a). The 2D 1 H-13 C HSQC spectrum shows resonances for all protons, which directly bound to 13 C (J1 coupling), with the chemical shift for 1 H in the xaxis and 13 C in the y-axis. Thus, the spectrum completely omits non-H-bonded 13 C atoms such as carbonyls in carboxylic acids. The H-C2(-OH) of 2-HG resonates at ( 1 H chemical shift, 13 C chemical shift) = (3.92 ppm, 72 ppm) and appears with a positive phase (brown in Fig. 1b). Only directly C-bonded H-atom contributes to this signal; thus, the proton on the oxygen remains occulted. The H-C4-H protons of 2-HG resonate at (2.157 ppm, 33 ppm) and appear in the negative phase (pink in Fig. 1b). Finally, the H-C3-H resonances appear as two distinct peaks with chemical shifts of (31 ppm, 1.73 ppm) and (31 ppm, 1.89 ppm) with a negative phase. From this 1 H-13 C HSQC spectrum standard, we also extracted the 1 H axis projections for each resonance in the 2D spectrum ( Supplementary   Fig. S2), which provides better visualization of the positive or negative phasing for each peak which, by extension, indicates the level of substitution at each corresponding carbon.
Having established a reference 2-HG HSQC spectrum, we then employed this pulse sequence to acquire the 1 H-13 C HSQC spectrum of live IDH1 mutant cancer cells (HT-1080) re-suspended in phosphate-buffered saline with 10% D2O in a standard 5 mm NMR tube with a 500 MHz NMR. HT-1080 cells harbor the R132C IDH1 mutation and overproduce 2-HG 27 , which we verified here by mass spec (Supplementary Fig. S1b). Additional to the 2-HG peaks, the spectrum shows peaks for other abundant metabolites such as lactate, myo-inositol, choline, fatty acids, and many others. To make sure we correctly assigned 2-HG peaks in the spectrum, and as the negative control, we scanned live HT-1080 cells treated with 10 µM mutant IDH1 inhibitor, AGI-5198 23 . The AGI-5198 inhibitor has an IC50 of 160 nM for R132C IDH1 mutant and IC50 of> 100 µM for IDH1 wildtype 23 . Therefore, treating IDH1 mutant cells with the non-toxic concentration of AGI-5198 inhibitor results in a significant reduction in 2-HG level, which was confirmed by doing mass-spectroscopy on our HT-1080 cells treated with this inhibitor (Supplementary Fig. S1b). Figure 1c shows the spectrum of live HT-1080 cells treated with 10 µM AGI-5198 for 48 hrs. Comparing the HT-1080 cells spectrum (Fig. 1b) with the spectrum of HT-1080 cells treated with AGI-5198 ( Fig. 1c) revealed that only one pair of peaks was sufficiently distinct from other metabolites to be diagnostically useful. Since and (1.89 ppm, 31ppm) of 2-HG are unique to IDH1 mutant cells, we also acquired the spectrum of live IDH1 wildtype cells (Fig 1d, Table 1 and Supplementary Fig. S3).
Comparing the spectrum of IDH1 wildtype with IDH1 mutant cells, we observed that H-  Table 1 shows the list of cells invitro and ex-vivo xenografted tumors that we acquired 1 H-13 C HSQC spectra on, the origin of cell lines, their IDH1 mutation status, and whether we detected H-C3-H peaks on their spectrum or not. To the best of our knowledge, this phase-sensitive HSQC pulse sequence has not been applied to biological experiments, and certainly not for diagnosing specific oncometabolites.
We looked at 1 H projection of 2D HSQC spectra to more robustly determined signal to noise ratios and quantitatively compare experimental conditions that alter levels of 2-HG detectable in this spectrum. However, these two peaks are absent in the spectra of freshly dissected IDH1 wildtype xenografted tumors ( Fig. 2b and Fig. 2c). Moreover, H-C3-H peaks are absent in the spectrum of the intact normal mouse brain (Fig. 2d).
We also looked at the spectra of intact IDH1 mutant and wildtype human GBMs (Fig. 3).
We confirmed the IDH1 mutation status, as well as high accumulation of 2-HG in the IDH1 mutant tumor by mass spectroscopy (Fig. 3b). Then, we looked at the spectrum of the intact IDH1 mutant human GBM, where we were able to detect H-C3-H peaks of 2-HG.
These two peaks were absent in the spectrum of the intact IDH1 wildtype human GBM.
Like cells spectra, the C2-H peak of 2-HG in IDH1 mutant tumors was convoluted to the Homozygous deletion of metabolic enzymes stands to cause accumulation of metabolites upstream of the deleted enzymes, and as such, be detectable by NMR based methods.
Having demonstrated that it is possible to genetically type a tumor based on metabolite levels by 1 H-13 C HSQC, we sought to expand its utility to additional oncometabolite aberrations such as those induced by passenger homozygous deletions of metabolic enzymes. A critical advantage of the 1 H-13 C HSQC spectrum is that the 2D peak patterns for specific metabolites are constellation-like and can serve as the fingerprint identification. The advantage of this compared to 1 H is that subtle changes in chemical shifts of specific molecules can happen due to biological microenvironmental changes, such as ionic strength and pH changes, which confuse the assignment of peaks to a specific molecule. With 1 H-13 C HSQC, the arrangement of peaks even shifted in 1D 1 H by microenvironmental effects are unlikely to affect the constellation pattern of peaks in the 2D 1 H-13 C HSQC. This advantage is perfectly illustrated in the detection of elevated gluconate in tumors with PGD-homozygous deletion. Figure 4a shows the oxidative pentose phosphate pathway (PPP) where glucose 6-phosphate produced from glycolysis enters into the oxidative arm of PPP and further converts to the 6-PG. PGD is the enzyme that is responsible for the oxidation of 6-PG to ribose 5-phosphate. The homozygous deletion of PGD results in the accumulation, i.e. 100-fold elevation of 6-PG and its hydrolysis product gluconate (Supplementary Fig. S4) compared to PGD-intact. Thus, the detection of 6-PG/gluconate can serve as the identification for tumors with this genomic alternation. Although both gluconate and 6-PG are elevated in the PGD-deleted cells, the concentration of gluconate is higher than 6-PG, which makes it suitable to detect with MRS. However, the detection of gluconate from the 1 H spectrum is challenging because all the protons of gluconate molecules are C-H, which is a common chemical group in all sugars and resonates very close to the water signal in 4.7 ppm (Supplementary Fig. S4). Therefore, peak patterning in the 2D 1 H-13 C HSQC is ideally suited with the detection of gluconate and typing the PGD-deleted tumors. In the Fig. 4b, the 1 H-13 C HSQC of gluconate standard with specific carbon hydrogen is shown using the phase sensitive 1 H-13 C HSQC in which the CH group appears in the positive phase, and the CH2 group appears in the negative phase. The C2-H peak of gluconate resonates at where the constellation of gluconate peaks are expected in the 2D spectrum, we overlaid tumor spectra with that of the gluconate chemical standard (black/gray color). Figure 4c shows the spectrum of intact PGD-deleted xenografted tumor (NB1), where we observed the near-perfect overlap of the peaks in both dimensions with gluconate standard peaks.
In contrast, Figure 4d shows the spectrum of the ex-vivo xenografted PGD-rescued tumor (NB1-PGD), which was derived from NB1 cells; the PGD deletion was rescued with the atopic expression of PGD and effectively generated the NB1 cell line which is PGD wildtype. All the peaks associated with gluconate are absent in the spectrum of the PGDrescued tumor. Figure 4e and 4f show 1 H-13 C HSQC spectrum of PGD-wildtype tumor (D-423) and normal mouse brain where except one peak of H-C6-H doublet, which is obscured by the (-CH2-) metabolites, the rest of C-Hs peaks of gluconate are absent in both spectra. Supplementary Figure S6 shows the 1 H projection of the gluconate standard, PGD-deleted, rescued and wildtype tumors from rows that we expect to observe gluconate peaks. Spectra of extracted rows of PGD deleted tumors show the same peaks as gluconate standard, while these peaks are absent in PGD wildtype tumors and normal brain. Altogether, we show that 1 H-13 C HSQC is a robust technique to type PGD deletion status of cells and tumors, in addition to IDH mutations and elevated 2-HG. This is illustrative of the fact that this technique can type multiple oncometabolic aberrations simultaneously and agnostically.

Discussion
Cancer metabolism remains an aspirational target for precision oncology 28 . Rapid identification of the subset of patients who are candidates for specific metabolism-based precision oncology drugs remains highly problematic. MRS has the potential to address this challenge, but the utility of this technique (as 1D 1 H MRS) is currently limited to the detection and quantification of unusually abundant metabolites, which are uninformative for the purposes outlined above. The 1D 1 H MRS sequence is employed routinely in the clinic; however, it suffers from broadening of peaks and lack of specificity 29 . For example, in the case of 2-HG detection, it is difficult to differentiate chemical shifts of metabolites like 2-HG from similarly structured metabolites such as glutamine and glutamate.
Employing the spectral editing technique can improve the convolution/specificity problem, but it results in a sensitivity decrease and loss of global metabolite overview 30 . The 13 C spectrum has better specificity/spectral resolution compared to the 1 H spectrum; however, it suffers from very low sensitivity due to the low natural abundance of 13 C (1.1%) 31 . The sensitivity problem can be addressed by using the 13 C hyperpolarized MRS 32,33 . While this technique is useful for very specific questions, its application is limited to the choice of the hyperpolarized probe, and hence metabolic pathways 21 .
Chaumeil et al used hyperpolarized 1-13 C ⍺-ketoglutarate to identify IDH mutant tumors non-invasively 34 . However, the technique's sensitivity is limited because of the presence of a 5-13 C ⍺-ketoglutarate peak, which has a chemical shift close to 1-13 C 2-HG 34 .
Moreover, the use of ⍺-ketoglutarate as the hyperpolarized probe can be restricted due to the hydrophilic nature of this molecule, which can limit its cell membrane permeability 35 .
Salamanca-Cardona et al also used the hyperpolarized 1-13 C glutamine to identify the IDH mutant tumors 36 . However, the transition of this technique to the clinic has not been forthcoming because of the short T1 half-life of 1-13 C glutamine 36 and slow glutamine uptake by cells 31 . Correlation spectroscopy (COSY) has been employed in-vivo setting to detect 2-HG 30,37 . While this method is capable of detecting 2-HG, its application does not extend to detecting metabolites such as gluconate, which has chemical shifts close to the 4.7 ppm water signal.
Having noted the limitations with other spectroscopic methods, in this paper, we present data to support the case for 2D 1 H-13 C HSQC in fulfilling the promises of MRS for precision oncology typing and metabolic profiling in vivo.
Specifically, we present proof-of-feasibility of metabolite profiling by 2D HSQC NMR in an intact biological setting ex-vivo. Compared to the 1D 1 H spectrum of cells, the 2D 1 H-13 C HSQC spectrum is more resolved, facilitating the assignment of specific peaks to specific metabolites since each 2D HSQC peak is defined by chemical shift information in both 1 H and 13 C axes. For example, in 2D 1 H-13 C HSQC spectra of tumors, one can easily differentiate the lactate peak from the broad mobile lipid peak at (1.2 ppm,30 ppm) which is elevated in high grade GBMs (necrosis) 29 , while these two peaks cannot be differentiated in 1D 1 H spectrum. We pioneer a phase-sensitive HSQC (HSQCETGPSISP3.2) pulse sequence to detect specific oncometabolite aberrations in biologically intact oncology samples for the purposes of precision oncology. The sequence not only yields a well-resolved 2D 1 H-13 C HSQC spectrum but also distinguishes CH2 from CH3/CH, providing an additional layer of information. For proofof-principle, we applied the technique to type tumors that are IDH1-mutant and characterized by detection of elevated 2-HG; to demonstrate the wider applicability to metabolic precision oncology, we further show that HSQC can readily identify tumors which are PGD-deleted, by detection of elevated 6-PG and gluconate. While here, we have demonstrated the utility of this technique for two specific metabolic aberrations caused by specific genomic alterations, this technique is applicable to the metabolomic typing of any other oncometabolite aberration whether genetic or epigenetic in nature.
While we see the present work as a stepping stone to the application of the phasesensitive 1 H-13 C HSQC sequence for precision oncology typing by MRS -and our data provide strong proof-of-principal for intact samples ex-vivo -legitimate questions may be raised as to whether 1 H-13 C HSQC could be carried out in a full in vivo setting. We reply to this concern by noting that, in fact, in-vivo 2D 1 H-13 C MRS method has already been pioneered by other groups 21,38-42 , though our study is the first to illustrate its utility for Besides its potential for IDH mutation typing in-vivo, the HSQC sequence developed here for ex-vivo samples could be applied essentially immediately in the (neuro)surgery setting to identify patients with IDH mutant tumors as well as delineating tumor versus normal brain tissue and in the future to type for other potentially targetable metabolic vulnerabilities as they are discovered. Because of the short NMR scan time and no requirement for sample preparation such as chemical extraction, this method would be well suited for the fast just-in-time surgery setting. The detailed metabolite information could complement other imaging and genetic tests such as in-vivo MRS as well as genomic sequencing and IHC. The cellular structure of tumors is preserved in this method, which can be used for further genomic and histopathological analysis. Compared to the NMR spectrum of polarly extracted tumor solution, which only detects polar metabolites, the spectrum of the intact tumor has information about polar and apolar metabolites. Indeed, 1 H-13 C HSQC has been employed to compliment 1 H metabolic profiling on tumor extracts using very high magnetic field and large amount of samples or on intact tumors using high resolution magic angle spinning 43,44 . We emphasize that the work here was performed in a standard NMR without signal enhancing features such as magic angle spinning and can be implemented in even non-specialist labs. In sum, 1 H-13 C HSQC stands to dramatically expand our knowledge of tumor metabolism in vivo and to help realize the potential of metabolic precision oncology. The spectra were analyzed using Bruker TopSpin 3.1. For the 1D 1 H spectra, we did 1 st and 2 nd order phase correction followed by automatic baseline correction. 2D 1 H-13 C HSQC spectra were phase-corrected as needed, and baseline adjusted at the F1 and F2 axis using abs1 and abs2 commands. Counter level increments were also adjusted to the 1.1, for a total of 28 levels.

Methods
The 1D 1 H projections of the 1 H-13 C HSQC spectra were extracted using the rsr command and specifying the desired row number. Signal to noise ratio (SNR) was obtained by defining the desired signal region by sigf1 and sigf2 commands on Topspin and then using the SINO command. To find the signal to noise ratio for negative peaks, we first adjusted the negative phased peaks to positive and then select region of interest using the SINO command.       mutant cell lines (in-vitro and in-vivo) that overproduce 2-HG due to this mutation. We were able to detect H-C3-H peaks of 2-HG in all these cell lines (in-vitro and in-vivo).

Polar
Then we treated them with the mutant IDH1 inhibitor (AGI-5198), which shows significant effects in reducing 2-HG production, and the H-C3-H peaks of 2-HG disappeared after the treatment. The spectrum of live IDH1 wildtype cells shows a complete absence of H-C3-H peaks of 2-HG (in-vitro and in-vivo). We also looked at the PGD-deleted cell line (NB1), which significantly accumulates gluconate due to this deletion. We were able to detect gluconate peaks in the spectrum of live cells and ex-vivo xenografted tumors.
These peaks are absent in the spectra of PGD-rescued and wildtype cells and tumors.    Figure 1