Noninvasive In Vivo Assessment of Cardiac Metabolism in the Healthy and Diabetic Human Heart Using Hyperpolarized 13C MRI

Supplemental Digital Content is available in the text.

endocardial borders were manually contoured, with subsequent analysis performed using cmr42 © (Circle Cardiovascular Imaging Inc, Calgary, Canada).
Myocardial 1 H magnetic resonance protocol A suitable mid-interventricular septal voxel was initially selected for assessment from long and short axis cardiac localizers. As previously described [38], myocardial lipid content was measured from a series of five water-suppressed, T1 and B1 insensitive Stimulated Echo Acquisition Mode (STEAM) spectra (TE 10 ms, TR 2 s, 5 averages), using a 6 channel flexible 1 H receive array. These were standardized using a single waterunsuppressed spectrum. All data were analysed in MATLAB (Natick, Massachusetts, USA) using an implementation of the AMARES algorithm [39]. The myocardial lipid content was calculated as a percentage relative to the water signal.
Myocardial 31 P magnetic resonance protocol As previously described in our centre [10,40], to minimize heart-coil distance, subjects were positioned prone on a dual-tuned commercially available 31 P/ 1 H surface coil (Siemens Healthineers, Erlangen, Germany). Long and short axis 1 H localizers were acquired to position the Chemical Shift Imaging (CSI) voxel grid. This was performed to ensure one voxel contained mid-septal myocardium at the mid-ventricular level.
Saturation bands were used to minimize signal contamination from skeletal muscle and liver. PCr/ATP was measured non-ECG-gated and free breathing, using a short echo time Hyperpolarized MR spectroscopy and data processing Subjects were scanned supine and short axis localizers, used to plan hyperpolarized data acquisition, were acquired using the inbuilt 1 H body coil. 13 C MR spectra were acquired using a 2 channel transmit, 8 channel surface receive array (Rapid Biomedical, Rimpar, Germany). A [ 13 C]urea fiducial marker strapped on top of the coil was used to calibrate the 13 C centre frequency. Spectra acquired at thermal equilibrium prior to hyperpolarized injection were used to subtract away the fiducial and background thermal-equilibrium fat signal from the hyperpolarized spectra. Hyperpolarized data were acquired from a mid-ventricular 10 mm axial slice, beginning at the start of the injection. To minimise motion, subjects were invited to initiate an end-inspirational breath-hold at the start of the injection, breathing when required after approximately 30 s, before repeating this procedure until the hyperpolarized signal had returned to thermal equilibrium. The MR acquisition was a pulse-acquire spectroscopy sequence with slice selection achieved with a 1.3 ms sinc excitation triggered by ECG-gating to the R-wave. A single excitation and spectroscopic readout was acquired every heartbeat and run for up to four minutes after injection (flip angle 10°, bandwidth 5 kHz, TR 500 ms, 2048 complex points). Multi-coil data were recombined in MATLAB using the Whitened Singular Value Decomposition algorithm [43], with coil combination weights calculated for spectra with the highest SNR subsequently applied to the entire dataset. Spectra were background-subtracted prior to quantification with the AMARES algorithm [39], with appropriate prior knowledge. Total integrated metabolite-to-pyruvate ratios, known to linearly correlate with first-order chemical kinetic rate constants [44], were calculated from 60 seconds of data taken after the initial appearance of the pyruvate resonance in the spectrum.

Statistical analysis
Linear mixed effects models form a statistically powerful generalisation of traditional ANOVA methods that are effectively able to use repeated measurements even in the presence of missing data, and are comparatively robust [45]. Accordingly, hyperpolarized datasets, quantified as described above, were analysed with the lme4 [46] and the car packages in R (v3.6.0, R Foundation for Statistical Computing, Vienna, Austria), with metabolic state and disease status considered as fixed effects, and subject ID considered as a random effect, and an ANOVA table computed. Data were subject to a Shapiro-Wilk normality test and one outlier corresponding to an unpaired fasted subject with diabetes with a Z-score of 9.4, was identified (Grubb's test p=.003, suggesting that point was an outlier) [47]. Data derived from this patient were excluded for subsequent analysis.
Unless otherwise stated, all other analyses were performed in GraphPad Prism (GraphPad Software, San Diego, California, USA) via simple unpaired unequal-variance ttests with the canonical p<0.05 threshold for statistical significance.