Introduction

Bipolar type I disorder (BD) is characterized by severe mood swings and occurs in about 1% of the population. The mechanisms underlying the disorder remain unknown. Prior studies have suggested abnormalities in brain metabolism using 1H and 31P magnetic resonance spectroscopy (MRS). Supporting altered metabolism, in previous studies we found T1ρ relaxation times in the cerebellum were elevated in participants with BD. In addition, T1ρ relaxation times in the basal ganglia were lower in participants with BD experiencing depressed mood. Based on these findings, this study sought to probe brain metabolism and extending these assessments to the cerebellum.

Methods

At the time of this analysis, 64 participants with BD and 42 controls completed both the 3T and 7T MR portions of the study and were included in the analysis. MRS data was collected on a GE 7T MR950 scanner. The 1H MRS data was collected using a NOVA 2Tx/32Rx channel while the 31P MRS data was collected using a dual tune RAPID Biomedical 31P /1H coil. For both the 1H and 31P based MRS measurements, an anatomical MP-RAGE was collected in the respective coils allowing for co-registration of the MRS data to anatomical data collected at 3T. For the 1H based measurements, single voxel data was collected from the cerebellar vermis and right putamen using a semi-LASER sequence. For the 31P MRS measurements, a 3D free induction decay (FID) sequence was used. Phosphocreatine (PCr) was set as the center frequency for the 31P measurements.

The T1 and T2 weighted scans collected at 3T were analyzed using BRAINS AutoWorkup. The 1H semi-LASER data for both the cerebellar vermis and right putamen were analyzed using LCModel with water as a reference standard. The coordinates of each of the 1H MRS voxels were used to generate binary masks for the 1H in the physical space of the T1-weighted anatomical image collected in the 2Tx/32Rx coil. This anatomical image was then aligned with the 3T T1-weighted scan using ANTs. The resulting transform was then applied to the binary masks. The binary masks were then used to measure the volume of gray matter, white matter, and cerebral spinal fluid (CSF) contained in each of the MRS voxels. The volume fractions of the tissue types were used to perform partial volume correction of the MRS data and generate molar metabolite concentrations as described by Near et al. (1) The 31P MRSI data were reconstructed and analyzed using in-house Python scripts. FIDs were generated from the k-space data by Fourier transformation in the spatial-dimensions. A 20 Hz exponential line broadening function was applied to the FIDs prior to Fourier transformation, phase and baseline correction. The peak amplitudes and peak positions were obtained from voxels with a peak SNR of PCr above 7 db. Intracellular pH (pHi) was estimated based on the chemical shift between PCr and inorganic phosphate (Pi). Metabolic images were generated for (pH, aATP, PCr) and upsampled using the pyramid_expand function of Scikit-learn.


The primary analysis investigated the difference in brain metabolite concentrations between participants with BD and controls using a series of multiple linear regression models where a separate model was estimated for each brain metabolite or imaging measure. Each model had the brain metabolite or imaging measure as the outcome and the predictors were diagnosis, age, and sex. False discovery rate correction was performed.

Results and Discussion

For the cerebellar vermis, the linear models for NAA, Glx, glutamate, glutamine, glutathione, taurine, choline, total creatine, aATP, and pHi were statistically significant (q-value£0.05) after correcting for multiple comparisons while myo-inositol and GABA had a q-value of less than 0.1. The models with significant omnibus q-values (i.e., q-value£0.05) revealed that diagnosis was a significant predictor for NAA, glutamate, glutathione, taurine, total creatine, and pHi indicating that these metabolites had significantly different concentrations in participants with BD as compared to controls. For the 1H metabolites, all of those with a significant diagnosis effect were higher in participants with BD. A reduction in pHi was found in participants with BD as compared to controls. For the putamen, none of the models had a q-value£0.05 although several metabolites did have an initial p-value£ 0.05 including Glx, glutamate, myo-inositol, choline, and pHi. Post-hoc analyses were performed for metabolites that showed a p-value£0.05 and were significantly different in the cerebellar vermis. This post-hoc analysis found that Glx, glutamate, choline, total creatine, and pHi showed a significant diagnosis effect indicating difference in the measures between participants with BD and controls.

Conclusion

The findings of reduced pHi in this study support models of mitochondrial dysfunction in BD where there is an impaired ability to utilize pyruvate in oxidative phosphorylation (2-3). A shift from oxidative phosphorylation toward glycolytic energy production likely increases lactate acid, CO2 levels, and free protons resulting in tissue acidosis. This shift in energy production may also increase reactive oxidative species leading to increased glutathione. These findings extend to the cerebellum.

Acknowledgements

This work is supported by NIH grants R01MH111578, S10OD025025, and S10RR028821.