Introduction

R2* is a sensitive marker for iron accumulation in the brain, and changes in T2*-weighted contrast have been reported in several neurodegenerative diseases^1,2,3. Iron accumulation in disease can also be cell-type specific^3,4. Magnetic resonance spectroscopy (MRS) offers the unique ability to investigate cell-specific properties that affect the MR signal. While water is ubiquitous in tissue, brain metabolites are almost exclusively found in the intracellular space and are preferential to cell types^5,6. Metabolite T2* could therefore reflect cell-type specific susceptibility changes in brain disorders, such as activation of glial cells associated with iron accumulation⁴. While the linewidth (FWHM) of a single resonance (i.e. creatine peak at 3.0ppm) can be used to estimate its T2*, the estimation of T2* for J-coupled systems (such as glutamate (Glu) and myo-inositol (Ins)) requires a more sophisticated approach. We propose a method that uses a single MRS acquisition to generate a set of time-shifted free induction decays (FIDs) from which metabolite T2* is estimated (fig.1). We measured T2* of five brain metabolites in a cohort of ten healthy volunteers, and studied reproducibility in three healthy volunteers.

Materials and methods

Experiments were conducted on a 7T whole-body MRI scanner (Philips Healthcare, The Netherlands) equipped with a volume transmit/32-channel receive head coil (Nova Medical, USA).
Data acquisition: In all experiments, 3D-T1W gradient-echo images (TR/TE=5/2ms, resolution=1x1x1mm³) were used for planning, and an 8mL volume-of-interest was positioned in the posterior cingulate cortex (PCC, fig.2) for the MRS experiment.
Cross-sectional study: Data were acquired in 10 heathy subjects (6F/4M) as part of data collection for the EUFIND consortium⁷. Metabolite and water spectra were acquired using sLASER (TR/TE=8000ms/34ms, NSA=32).
Reproducibility assessment: Repeated data were collected from 3 heathy subjects (1F/2M, n_subject1=6/ n_subject2=4/n_subject3=5). Metabolite and water spectra were acquired from the same location in the PCC with sLASER (TR/TE=6000ms/29ms, NSA=48) with different shim settings on different days.
T2* calculation: The process is illustrated in fig.1. A set of FIDs with different time-shifts was generated by progressively discarding the first points of the originally acquired FID. Seven spectra were thus generated, with shift values between 0ms and 30ms in 5ms intervals. The resulting spectra were quantified using LCModel⁸. A basis-set was simulated for each time-shift, accounting for any J-evolution and the first-order phase variation across the spectrum induced by the time-shift. To estimate metabolite T2*, a linear regression was performed to fit the logarithm of the metabolite signal as a function of the time-shift. The water and total creatine (tCr) resonance linewidths (FWHM_water/FWHM_tCr) were measured with an in-house Matlab® linear prediction and singular value decomposition (LPSVD) routine.
Statistical analysis: Statistical significance was tested using GraphPad Prism 7 (GraphPad Software, USA) using a paired Student’s t-test.

Results and discussion

The LCModel fits to the time-shifted data were excellent for five brain metabolites (N-acetyl-aspartate (NAA), Glu, Ins, total choline (tCho) and tCr, fig.1B), showing that the basis-sets fully accounted for any J-evolution and first-order phase evolution caused by the time-shift.
In our cross-sectional experiment, the shim was relatively constant (FWHM_water~10.2-11.5Hz), LCModel Cramer-Rao lower bounds for all metabolites and all time-shifts < 15, and the variance in metabolite T2* values across subjects was low. As illustrated in fig.2B/C, the ln(tCr) decreased linearly with time-shift (R²=0.98, p<0.0001) and the FWHM_tCr estimated from the T2* value was linearly correlated with the one measured with LPSVD (R²=0.89, p<0.0001). This suggests that T2* of brain metabolites can be reliably obtained in a simple manner with similar shim values across subjects. As shown in fig.3A, the linear regression of the ln(metabolite signal) with respect to the time-shift was also highly significant for all other metabolites (R²>0.89, p<0.001). These tight fits to a mono-exponential T2*, together with small variations in B0 homogeneity resulted in significant differences in T2* across metabolites (fig.3B). These differences can be attributed to a combination of intrinsic T2 differences across metabolites⁹ and local susceptibility differences across cell types.
In the reproducibility study, we intentionally introduced a larger range of shim values across measurements, resulting in FWHM_water values between 10 and 15Hz. Differences in B0 homogeneity strongly affected the variance in T2* estimates within subjects (fig.4A). Strong correlation between metabolite and water T2* (fig.4B) suggested that the FWHM_water could be used to account for some of the variance introduced by differences in B0 homogeneity. Fig.4C shows the T2* values of the same metabolites normalized by the water T2*. The variance of normalized metabolite T2* values within subject significantly decreased (fig.4D).

Conclusion

We showed that the T2* of brain metabolites can be reliably obtained in a simple manner from a set of time-shifted FIDs originating from a single MRS dataset. With constant shim values across subjects, T2* values were significantly different across metabolites within a small cohort. As shown in our reproducibility study, normalization to FWHM_water can be partly used to account for the variance of T2* values due to shim differences across acquisitions. However, since the FWHM_water depends also on local tissue susceptibility, we are investigating the possibility of using B0 maps to account exclusively for shim differences. We envision a variety of applications for this method, including studying cell-specific changes in T2* in disease and assessing the BOLD effect on metabolite T2* during a functional task¹⁰.

Acknowledgements

This project has received funding from the Leiden University Fund (W-19356-2-32), and LEaDing Fellows COFUND programme. We thank Drs. D. Deelchand and P.G. Henry from Center of Magnetic Resonance Research at University of Minnesota (USA) and Dr. Julien Valette at Atomic Energy and Alternative Energies Commission in Paris (France) for sharing their Matlab programs to create LCModel basis-sets. We thank the European Ultrahigh-Field Imaging Network for Neurodegenerative Diseases (EUFIND) for letting us use the MRS data we acquired for the consortium.

References

[1] S. Rooden et al., Alzheimers Dement. (2014); [2] J.M.G. van Bergen et al., Am. J. Neuroradiol. (2016); [3] M. Bulk et al., Neuroimage Clin. (2020); [4] K.M. Gillen et al., Front Immunol. (2018); [5] Urenjak et al., J. Neurosci. (1993); [6] Choi et al., NMR Biomed. (2007); [7] E. Düzel et al., J. DADM. (2019); [8] Provencher et al., MRM (1993); [9] M. Malgorzata et al. NMR Biomed. (2012); [10] Bednařík et al., J. Cereb. Blood Flow Metab. (2015)