Introduction

Metabolic reprogramming, also known as the Warburg Effect in which lactate production via anerobic glycolysis (GLY) is favored over oxidative phosphorylation (OXPHOS), is a hallmark of cancer¹. This effect is particularly pronounced in glioma and other brain tumors and is the target of multiple oncologic therapies². However, limitations on measuring this abnormal tumor metabolism remains a crucial obstacle to clinical translation.

The recent introduction of deuterium metabolic imaging (DMI) offers a promising approach for in vivo imaging of the Warburg effect¹. Work by de Feyter et al. showed DMI can map the balance of glycolysis and mitochondrial metabolism in patiets with glioma at 4T 60-75 min following oral consumption of deuterated glucose ([6,6’-²H₂]Glc)³. Analogous to the ¹³C Lac/Bic ratio, the ratio of ²H-Lac/Glx (glutamate+glutamine) provides a direct measure of the Warburg effect. In a pre-clinical animal model study, Kreis, et. al. further showed that dynamic DMI acquisitions could also be used to acquire quantitative and spatially resolved measurements of glycolytic flux in tumors, which can be used to assess treatment response⁴. Although much slower than HP¹³C (~2 vs 40 min acquisition), DMI requires no FDA Investigational New Drug approval, no intravenous injection, no spin polarizer (a ~$2 million instrument with an associated yearly service contract) and uses less expensive materials (~$600/study for human [6,6’-²H₂]Glc vs ~$3k/injection for HP¹³C). In this project, we sought to implement the needed hardware for DMI on a clinical 3T PET/MR scanner (GE Healthcare, Waukesha, WI) and demonstrate high quality metabolic maps of HDO and ²H-Glx in adult brains.

Methods

We first constructed a 3T ²H birdcage coil. The coil used a lowpass quadrature T/R birdcage head coil design built from high conductivity .25” copper tubing and was tuned to 19.6 MHz and fed with quarter wave feed lines. This yielded an unloaded Q of greater than 300. We then used a commercial deuterium coil interface by Clinical MR Solutions for ¹³C that was re-tuned from 32.12 MHz down to 19.6 MHz. The interface also contained a narrow band TR switch, low noise preamp, and a quadrature splitter combiner.

On our 3T PET MRI GE scanner, the gradient drivers conduct harmonics into the 19-20 MHz band that are not adequately filtered using the standard gradient filters (consisting of common mode ferrite cores and capacitors to ground) mounted on the penetration panel. To remedy this problem, the filter was modified so that half of the cores were differential and the capacitors were increased 8-fold (Figure 1).

Using the modified gradient filter and ²H volume RF coil, we conducted initial investigations of in vivo whole-brain DMI in healthy adult volunteers following the oral ingestion of 60g/kg of deuterated glucose ([6,6’-²H₂]Glc). DMI data was acquired with both conventional and spherical k-space encoded chemical shift imaging (CSI) and a 90° non-selective RF excitation pulse. A 10x10x10 cm spatial array with 2.4 cm isotropic resolution was acquired with a repetition time (TR) of 350 ms, yielding a 5.8 min. and 2.9 min. minimum scan time for cubic and spherically encoded 3D CSI respectively. The high-quality spectra demonstrated in the next section used 8 averages for a total acquisition time of ~23 minutes. An additional T₁-weighted ¹H image was acquired for anatomical comparison and co-registration.

Results & Discussion

As shown in Figure 2, the modified gradient filter successfully removed the excess noise on our scanner seen in the 19-20 MHz RF frequency band. This led to good spectral quality across the brain volume, achieving a signal-to-noise ratio (SNR) of ~33 for HDO and ~5 for ²H-Glx at a representative center voxel spectrum (Figure 3). Note that the SNR for the ²H-Lac peak was too low in normal healthy volunteers to quantify reliably. Metabolic maps (computed via simple peak integration) of HDO and ²H-Glx demonstrate good anatomical correspondence with the T₁-weighted ¹H scans (Figure 4). In particular, the elevated HDO and reduced ²H-Glx signal in the center of the brain volume correspond well with the cerebrospinal fluid (CSF) in the proton scan. We additionally demonstrate that spherical k-space sampling yields an SNR gain of approximately √2 over cubic k-space sampling due to the twice the number of acquisition averages (Figure 5).

Conclusion

Following the oral ingestion of deuterated glucose, we used a modified gradient filter, home-built ²H volume RF coil, and spherical k-space sampling in a 3D CSI acquisition to obtain high quality adult brain metabolic maps of ²H-labeled water and ²H-Glx products of glucose metabolism. Future work will include the construction of an improved ²H head coil and an assessment of image quality improvements achievable by exploiting the mutual information content among DMI, MRI, and FDG-PET acquisitions. It is also important to note that not all clinical 3T MRI scanners suffer from the same gradient noise problem seen on our GE 3T PET/MR system. For these scanners, no gradient filter modifications would be required.