Hyperpolarized ¹³C metabolic imaging allows in vivo measurements of enzymatic activity, which provides valuable information for assessing tumor aggressiveness and treatment response. Single-time point Magnetic Resonance Spectroscopic Imaging (MRSI) enables the measurement of metabolite ratios, but is prone to scan timing errors and provides less information than dynamic acquisition methods. The goal of this project was to compare two dynamic sequences and their corresponding measurements of the metabolic conversion rate of pyruvate to lactate (K_PL) in cancer.

3D Dynamic MRSI: This 3D dynamic compressed-sensing ¹³C-MRSI sequence included a multiband excitation pulse for efficient magnetization usage, two adiabatic pulses for B1-insensitive spin-echo refocusing, and an EPSI readout gradient with random phase-encode blips for rapid data acquisition¹. The sequence was optimized for efficient SNR for kinetic modeling, with a spatial resolution of 3.3 x 3.3 x 5.4 mm³ and a temporal resolution of 2 seconds.

Rapid Exchange Spectroscopy (MAD-STEAM): The slab-selective dynamic Metabolic Activity Decomposition Stimulated-Echo Acquisition Method (MAD-STEAM) separates newly converted metabolites from existing ones based on the phase information, allowing for rapid quantification of magnetization exchange and metabolic conversion. It included an encoding portion compromised of two 90-degree slice selective pulses, a progressive flip angle scheme for efficient magnetization usage, and two adiabatic 180 pulses for phase sensitive reconstruction². The slab thickness was prescribed based on the tumor size.

Hyperpolarization and Imaging: [1-¹³C] pyruvate and ¹³C urea were co-polarized in a 3.35T GE SpinLab polarizer for 2 hours. After dissolution, 350uL of the hyperpolarized substrates were injected into the mice for a period of 15 seconds through tail vein catheters. The 3D dynamic compressed-sensing sequence started right after the injection, and the MAD-STEAM sequence acquisition started after 10 seconds of injection completion. The two hyperpolarized injections were performed on the same animal during a one-hour interval. The transgenic adenocarcinoma of mouse prostate (TRAMP) murine model was studied in the imaging sessions, and imaging was performed on a 3T clinical MRI system (GE, Waukesha, WI, USA) with a ¹H-¹³C dual tuned, birdcage coil. All animal studies were approved by the Institutional Animal Care and Use Committee.

Metabolic modeling: The area under the peaks were integrated for kinetic modeling. The 3D dynamic MRSI used a kinetic model where all metabolites were assumed to have the same T1 relaxation rate and lactate to pyruvate and alanine to pyruvate conversion rates were neglected³, as shown in Figure 1. These assumptions are required for robust fitting with this type of acquisition. For the slab-selective MAD-STEAM acquisition, the pre-existing and newly converted metabolites were separated based on the real and imaginary spectra respectively. This allows for use of a more complete kinetic model where both pyruvate and lactate T1 relaxation and bi-directional conversions were considered, as shown in Figure 2.

Three out of the 4 TRAMP tumors studied were solid tumors with diameters of over 1.5cm. One was early-stage hyperplasia, and with a diameter under 1cm. Both sequences and their corresponding reconstruction results confirmed that the early stage hyperplasia showed low tumor K_PL values compare to the other 3 TRAMP tumors. In addition, the two measurement methods showed very similar pyruvate to lactate conversion rates in the same animal as shown in Figure 3. A previous study showed that K_PL values measured using the 3D dynamic compressed sensing sequence significantly correlated with histological grade⁴. From this study, the similarity between the measured K_PL values from the two sequences suggests that the MAD-STEAM approach could also provide a valuable estimate of prostate cancer grade. We have previously shown that kinetic modeling based on the MAD-STEAM approach provides more accurate estimates of metabolic conversion rates⁵. This results from the separation of the pre-existing and newly converted metabolites, which allows use of a more accurate kinetic model. The 3D compressed sensing approach provides higher SNR and better spatial localization, but more assumptions are applied during the modeling process. We found a good correlation between the two methods, but expect MAD-STEAM to provide a stronger correlation with the underlying LDH enzymatic activity. By applying the two dynamic sequences on the same animal in sequential injections, we demonstrated that the two sequences provide similar enzymatic conversion rates. Both sequences studied in this project could be powerful tools for in vivo assessment of tumor stage, aggressiveness, and treatment response.