Synopsis
Neurochemical profiling
of rat or mouse brain by MRS and MRSI requires optimization of many steps.
Despite a strong magnetic field and the latest RF coil technology available,
the spectral quality obtained might not be as expected. This presentation will
give an overview of other factors that could be considered to achieve a
consistent high quality spectroscopic dataset. Target Audience
- Familiar with basic in vivo MR spectroscopy either on clinical systems or on preclinical systems
- Familiar with the basic concept of single voxel spectroscopy and chemical shift imaging.
- Works with small animal MRS and aims to identify sources that can improve the final spectral quality
Objective
- Understand the difference between single voxel MR spectroscopy and spectroscopic imaging and its use in small animal in vivo spectroscopy.
- Identify elements in the experimental in vivo MRS/MRSI protocol that will lead to significant improvement in the spectra quality.
- Define some minimal quality criteria for performing neurochemical profiling for in vivo small animal MRS/MRSI
Purpose
MR spectroscopy has been used since the 70s ([1, 2]) to study the metabolite changes/modulations in different physiological and pathological conditions in vivo. It is important to realize that it is challenging to study the rodent brain on the same level of detail in vivo as can be done with MRI in the human brain. Due to the size difference between human and rodent brain (see figure 1, concept adapted from [3], human brain image from [4]), voxel volumes have to be reduced by a factor of 103-106 compared to the human brain [3]. Apart from the size difference, the curvature of the skull in rodents makes it harder to shim well over the whole brain. Air-tissue interfaces e.g., between the auditory cavities and the brain create complex, strong, and localized magnetic field distortions [5] that effect the mouse brain more due its smaller size compared to the rat brain. The development of the MR technology by producing scanners that operate at higher magnetic field strengths, have better hardware, improved RF technology and RF coil design and by improved pulse sequences led to better possibilities to visualize the anatomy and metabolism in the animal.
Classically, two classes of spectroscopy measurements can be performed, acquisition of single spectra (MRS) and so-called spectroscopic imaging (MRSI or CSI). In this lecture the purpose is determine the quality requirements to perform a so-called neurochemical profiling at short enough echo times to detect as many metabolites as possible in a region of interest in the rat or mouse brain.
Single voxel MR spectroscopy (MRS)
Tissue specific localization can be obtained by using a small RF-coil on the object of interest [6] and will work well for the acquisition of 31P-MRS of muscle tissue as the 31P-signals of interest are muscle specific (see for example [7]). But for most, if not all, 1H-MRS experiments, a localization sequence like STEAM [8], PRESS [9], LASER (ref [10]), SPECIAL ([11]), semi-Laser ([12]) is needed in order to be able to acquire relevant data. If neurochemical profiling of the the brain tissue [13-17] is the main focus, a short echo time of below 10 ms is crucial in order to be able the determine the metabolites that have a short T2* due to j-coupling. An echo time below 10 ms cannot be reached with some of the single voxel sequences mentioned above. Depending on the minimal echo time required, the type of RF coil available and the type of sequence available on the MR platform, one of the pulses sequences above can be choosen. The preferred RF-coil for MRS data acquisition in terms of sensitivity is a small transmit-receive surface coil positioned on the area of interest. It might be possible to use a volume coil for transmit and a surface coil for the receive path. The availability of cryogenic RF coils [18] pushes the limits even further by improving the SNR for 1H-MRI/MRS with a factor of around 2-2.5 compared to a surface coil; keep in mind that the investment for such a coil is about 100 times higher than a simple transmit-receive surface coil. The advantage of volume coils for transmit lies in a homogenous B1-excitation profile, but an efficient working volume RF coils is needed in order to obtain enough B1 (power) for the high-bandwidth RF pulses that are needed to minimize the chemical shift displacement artifact (for example [19]). The chemical shift displacement artifact is an important parameter but in this lecture it is assumed that the sequences available on the MR scanner are optimized for this to minimize this artifact.
A good size surface RF coil for MRS experiments in the the mouse brain would be a 15x11 mm2 surface coil as it will fit nicely around the skull at the level of the occipital and frontal bone. It is not a good choice for MR imaging to perform global structural analysis of the brain due to typical inhomogeneous B1 profile of surface coils; the 90/180 degree RF pulse is only defined in a relative small area given a certain power setting. The problem of the inhomogeneous B1-field can be partly overcome by using adiabatic pulses [10].
In order to achieve a complete neurochemical profiling of the rodent brain [15], a short echo time and relatively long repition time (4-5 s) is needed to detect around 20 metabolites. Both the STEAM [16, 17] and SPECIAL [14, 20] pulse sequences have been proven to be useful. The SPECIAL sequence achieves about twice the signal over the STEAM sequence, but as the SPECIAL employs a subtraction technique, it is more sensitive to movement.
Fixation of the animal is crucial to minimize artifacts caused by movement. It is important that the animal is breathing freely. To much pressure on the animal and the animal will not survive. But even in the intermediate regime where fixation looks okay, too much pressure can cause the animal to gasp after some time. This results in a lower breathing frequency (30/50) for mice) accompanied by large movements of the animal. It would be better to aim for a higher breathing frequency (70-90) with small respiration amplitudes to get better quality spectra.
In our lab, we choose the SPECIAL sequence combined with outer volume suppression (OVS) for all our MRS experiments to improve the localizing performance and reduce the demand for crusher gradients in the localization sequence [21]. Our implementation of the SPECIAL sequence (kindly provided by EPFL, Lausanne) doesn’t allow for oblique voxels to guarantee that the adiabatic RF pulse designed to cope with the inhomogeneous B1-profile is in the vertical direction perpendicular to the plane of the RF coil. It is therefor important that the animal is positioned correctly in the magnet and the brain is not tilted. This can be interpreted as a limitation, but it turns out that this helps to make the experiments more robust and reproducible in terms of shimming and RF-pulse calibration by limiting the possibility of rotating the voxel. The use of the SPECIAL sequence in combination with a volume transmit coil is not recommended as it it requires an efficient working volume coil to obtain enough B1 to fulfill the adiabatic condition of the RF pulse. This can be easily checked by plotting the water signal as a function of the 90 degree RF pulse power. The adiabatic condition is reached when a plateau phase is reached. When using a large coil, STEAM might be easier to use as the 90-degree pulse needs less power.
Like any MRI experiment, MRS voxel sizes are a comprise between SNR/scan time and the specific localization needed in the tissue. Volumes can become so small that no good spectrum can be acquired in the time frame available for animal experiments. Animals are pretty stable using isoflurane anesthesia during the first 1.5-2 hours. After that time period, often breathing rates suddenly goes down and adjustments of the isoflurane dose has to be done. As the animal is less stable, the quality of the MR spectra will be hampered as well. One should be aware that the isoflurane effects the metabolite levels measured in the mouse brain, especially lactate levels [22].
The quality of MR spectra is described in terms of SNR and the linewidth of the metabolite signals. The quality of the shimming procedure can be considered as a crucial factor for the end-users as this will change from animal to animal. Shimming can be done manually although it is highly recommended to use an automated method using for example FASTMAP [23] or phase map shimming (for principle see [19]). First and second order shims are both prerequisites to achieve a reasonable shim [17]. To obtain a dataset with consistent quality, it is important to define quality criteria for the shimming. The volume to be shimmed is a bit larger, typical 0.1-0.2 mm, than the voxel that should be measured. For a magnetic field strength of 9.4T using a SPECIAL sequence with an echo time of 2.8 ms, a good shim quality is a water linewidth of 15 Hz (FWHM) for mouse brain and 18 Hz for rat brain spectroscopy [13]. By experience, it is observed that the linewidth of the water line tends to increase during scanning; the final linewidth should not exceed 20-21 Hz at 9.4T, otherwise the dataset should be discarded. If the predefined shim quality cannot be reached, reposition the volume and try again. If possible the voxel size can be slightly decreased. It is recommended to terminate the MRS measurement if the quality criteria cannot be reached. The animal is removed from the magnet and depending on the study design will be measured again or the data of that animal will not be included at all.
The second crucial optimization that has to be carried out for every experiment is the water suppression. Although even here different methods exist, and at the end the quality of the water suppression that can be reached is most important, VAPOR water suppression [21] is a good method to use if available on the scanner. An almost perfect water suppression can be obtained by optimizing the strength of the water suppression pulses and the delay of the last RF water suppression pulse.
Even though the software of the sequence would allow to choose small volumes of 1 mm3, it is not directly obvious that the actual volume measured matches the volume chosen as RF-pulses might not be optimized in that region. Contribution of signals outside the volume chose will get relatively larger and the demand on the quality of the outer volume suppression (OVS) increases. When unexpected results occur when going to smaller volumes, a first check will be to check the sequence in a phantom to see if the signal scales with the volume selected. Some software packages provide automatic recalculation of the RF pulse to match the requirement of the voxel size selected.
MRS measurements typically take 20-40 minutes for good brain MRS spectra of small volumes. Often, the good shim obtained prior to the scanning might not be reflected into the spectral resolution of the final dataset. As the animal might move, subtle changes in body temperature during scanning, changes in breathing pattern and changes in anaesthesia levels all might cause small drifts in the frequency of the spectrum. Scanning in blocks will help to correct for this. One larger movement of the animal will allow you to simply discard the corrupt block. Although one could save every single spectrum during scanning, 16 scans seem to be a nice compromise for amount of data to be handled. One single block takes approximately one minute of scan time; 16 scans will give you even the advantage of phase cycling. The improved SNR of the data after 16 scans compared to single acquisition scans even facilitates the frequency correction of the data.
Frequency correction is easily done in spectra; one could identify the creatine peak at 3.03 ppm and estimate the shift needed for alignment between all blocks.
- As frequency correction is easily performed in the frequency domain, the frequency shift is estimated in the spectra and this frequency shift is applied to the time domain data. MRS analysis programs like jMRUI [24-26] and LCModel [27] expect time domain data as input. jMRUI performs the quantitation in the time domain while LCModel uses frequency domain for its analysis.
- The frequency correction needed is often smaller than spectral resolution of your spectrum. It helps to Fourier transform the data with a factor of 4 to interpolate between these points. This interpolation helps to get better alignment.
- Make sure to correctly handle the amplitudes of the spectra in order not to effect the quantification values during this pre-processing step.
The result of this frequency alignment is shown in figure 2. The linewidth is for this improved with 20% and is more in agreement with the expected linewidth based on the quality of the shim.
Spectroscopic imaging (MRSI) / chemical shift imaging (CSI)
Magnetic resonance spectroscopic imaging (MRSI) or chemical shift imaging (CSI) is a technique to map metabolites from multiple locations simultaneously. MRSI can visualize spatial variations in a tissue slice or even in 3D. The result of the measurement is a grid of spectra covering the area of interest, e.g. a 16x16 CSI will give you 256 spectra over a volume. For every spatial dimension, an equal amount of phase steps has to be applied; in the case of the 16x16 CSI, 256 repetitions are needed when using a repetition time of about 4 s and will take almost 17 minutes. Seventeen minutes is a long time to get one complete coverage of k-space but one should realize that the whole CSI volume is measured 256 times. The SNR in a CSI volume element is about the same signal per volume as a 256 scan single volume MRS scan [19].
One could argue that CSI is the method of choice for scans that require more averaging, but there are some differences between the CSI and single voxel scans that complicate the spectroscopic imaging experiment. For example, a STEAM SVS experiment with voxel size of 2x2x2 mm3 would require more than 64 scans to get a decent SNR. As the mouse brain is about 10 mm wide, a CSI with a 16 x 16 mm2 field of view would cover the whole mouse head. It is important that the field of view (FOV) is larger than the object to prevent aliasing extraneous signals, also for CSI. The 8x8 CSI to achieve this would theoretically give the same SNR in 64 scans as the STEAM measurement for one location. The problem is that the 8x8 CSI approach leads to problems and invalid spectra if no extra measures are taken; one of the most important ones is the effect of the point spread function on the spectral content.
The point spread function (PSF) describes the ‘leakage’ of one voxel in the CSI to another voxel, or in other words, it describes how much of other voxels in the CSI contribute to the one voxel you consider. The effect of the point spread function can be simulated for a 1D-CSI case (figure 3). Considering three voxels with a signal amplitude, two of them with amplitude 10 and one with amplitude 1. The expected profile in 1D can be plotted and would show in the ideal case a representation of the three volumes with amplitudes 10:1:10. Increasing the amount of phase steps will lead to a higher spatial resolution. The effect of the PSF causes voxels to contribute to other positions; the effect at 8 phase encoding steps is almost a horizontal line, no information can be obtained (see figure 3, second row). The measured spectral pattern improves at 16 phase encoding steps, but in this simulation performed in figure 3, 32 steps are needed to resolve the voxel in the middle with amplitude 1.
Although the simulation in figure 3 shows that 32 phase encoding steps are better, it is not something that can easily be applied in rodent spectroscopic imaging. The voxel size gets really small for a field of view of 20x20 mm2 (around 0.6x0.6 mm2). These volumes would probably be too small to achieve a good SNR per spectrum. Furthermore, 1024 phase encoding steps are needed for full k-space sampling and with a repetition time of 4 s takes more than 1 hour to complete. Given the small voxel of 0.6x0.6 mm2 in this example, 1 average might even not be enough to get a usable SNR. Long scan times are not really recommendable for animal imaging; 1.5 hour would be the upper limit with stable anesthesia on animals that are healthy (but preferable less than 45 minutes).
There are some ways to improve the PSF function. A Hanning [28] or a Hamming filter [29] improves the PSF significantly. The filter effect can be obtained by scanning the center of k-space more dense than the higher frequencies. This would require multiple averaging (up to 10 in the center of k-space [19] and is not feasible to perform unless a shorter repetition time (<2 s) is chosen. Filtering can also be applied during the reconstruction of the data. A Hamming filter will improve the PSF dramatically as shown in figure 3; the voxel with amplitude 1 can be resolved at a 16 phase step resolution. Even the 8 step improves, but in practice, 8 phase steps in a CSI should be avoided as the signal content in every voxel is contaminated with other positions to a large extent. The concept of the PSF becomes even more relevant in vivo where lipids are present outside the brain. In that case the ratio 10:1 could be more like 1000:1; even the smallest contribution from the lipids to the middle position would effect the spectra. For that reason, lipid signals should be suppressed and the easiest way would be to apply outer volume suppression to get down to the 10:1 ratio regime.
The larger volume to be scanned in spectroscopic imaging also complicates the shimming procedure; it is simply not possible to get the same kind of shim quality over the whole brain as in one single small volume. Furthermore, most shimming routines require a rectangular volume to be shimmed; skull should be avoided and only part of the brain can be shimmed (some MR software supports non-rectangular volumes though). More advanced methods are available to reach a high quality shim over a larger volume and these methods are superior in performance but they are not standard available, neither easy to implement [5].
In the case the central area of the brain is of interest, one could combine single voxel techniques with CSI encoding [19]; the resulting dataset is that only CSI voxels that are inside the single voxel preselection volume have signal, the other voxels only contain noise. These techniques for spectroscopic imaging can be called for example STEAM-CSI (for example [30]), PRESS-CSI (for example [31]), SPECIAL-CSI (for example [32]) or semi-LASER-CSI (for example [33]) can be used. This will help to get a high quality shim over a smaller volume, avoiding lipid contamination and at the same time keep the PSF effect in the CSI to a minimal.
Scan techniques exist to speed up the CSI acquisition, but they all have some kind of penalty to the spectral quality. A good starting point for different high speed CSI can be found in [19].
Acknowledgements
The author thanks Sandra Cuellar-Baena and Michael Gottschalk for valuable contributions to and proof reading of the manuscript.References
1. Gadian, D.G., et al., Phosphorus nuclear magnetic resonance studies on normoxic and ischemic cardiac tissue. Proc Natl Acad Sci U S A, 1976. 73(12): p. 4446-8.
2. Hoult, D.I., et al., Observation of tissue metabolites using 31P nuclear magnetic resonance. Nature, 1974. 252(5481): p. 285-7.
3. Rudin, M., Molecular imaging : basic principles and applications in biomedical research. Second edition. ed. 2013, London: Imperial College Press. xxiv, 593 pages.
4. By KieranMaher at English Wikibooks [Public domain], via Wikimedia Commons.
5. Juchem, C., et al., Multicoil shimming of the mouse brain. Magn Reson Med, 2011. 66(3): p. 893-900.
6. Bottomley, P.A., et al., Spatial localization in 31P and 13C NMR spectroscopy in vivo using surface coils. Magn Reson Med, 1984. 1(3): p. 410-3.
7. Heerschap, A., et al., Introduction to in vivo 31P magnetic resonance spectroscopy of (human) skeletal muscle. Proc Nutr Soc, 1999. 58(4): p. 861-70.
8. Frahm, J., et al., Localized high-resolution proton NMR spectroscopy using stimulated echoes: initial applications to human brain in vivo. Magn Reson Med, 1989. 9(1): p. 79-93.
9. Bottomley, P.A., Spatial localization in NMR spectroscopy in vivo. Ann N Y Acad Sci, 1987. 508: p. 333-48.
10. Garwood, M. and L. DelaBarre, The return of the frequency sweep: designing adiabatic pulses for contemporary NMR. J Magn Reson, 2001. 153(2): p. 155-77.
11. Mlynarik, V., et al., Localized short-echo-time proton MR spectroscopy with full signal-intensity acquisition. Magn Reson Med, 2006. 56(5): p. 965-70.
12. Scheenen, T.W., A. Heerschap, and D.W. Klomp, Towards 1H-MRSI of the human brain at 7T with slice-selective adiabatic refocusing pulses. MAGMA, 2008. 21(1-2): p. 95-101.
13. Cuellar-Baena, S., et al., Assessment of brain metabolite correlates of AAV-mediated overexpression of human alpha-synuclein in cortical neurons by in vivo H-MR spectroscopy at 9.4 T. J Neurochem, 2016.
14. Lei, H., et al., Neurochemical profile of the mouse hypothalamus using in vivo 1H MRS at 14.1T. NMR Biomed, 2010. 23(6): p. 578-83.
15. Pfeuffer, J., et al., Toward an in vivo neurochemical profile: quantification of 18 metabolites in short-echo-time (1)H NMR spectra of the rat brain. J Magn Reson, 1999. 141(1): p. 104-20.
16. Tkac, I., et al., Neurochemical changes in Huntington R6/2 mouse striatum detected by in vivo 1H NMR spectroscopy. J Neurochem, 2007. 100(5): p. 1397-406.
17. Tkac, I., et al., Highly resolved in vivo 1H NMR spectroscopy of the mouse brain at 9.4 T. Magn Reson Med, 2004. 52(3): p. 478-84.
18. Baltes, C., et al., Micro MRI of the mouse brain using a novel 400 MHz cryogenic quadrature RF probe. NMR Biomed, 2009. 22(8): p. 834-42.
19. De Graaf, R.A., In vivo NMR spectroscopy : principles and techniques. 2nd ed. 2007, Chichester, West Sussex, England ; Hoboken, NJ: John Wiley & Sons. xxi, 570 p., 8 p. of plates.
20. Cudalbu, C., et al., In Vivo Longitudinal (1)H MRS Study of Transgenic Mouse Models of Prion Disease in the Hippocampus and Cerebellum at 14.1 T. Neurochem Res, 2015. 40(12): p. 2639-46.
21. Tkac, I., et al., In vivo 1H NMR spectroscopy of rat brain at 1 ms echo time. Magn Reson Med, 1999. 41(4): p. 649-56.
22. Boretius, S., et al., Halogenated volatile anesthetics alter brain metabolism as revealed by proton magnetic resonance spectroscopy of mice in vivo. Neuroimage, 2013. 69: p. 244-55.
23. Gruetter, R., Automatic, localized in vivo adjustment of all first- and second-order shim coils. Magn Reson Med, 1993. 29(6): p. 804-11.
24. Naressi, A., et al., Java-based graphical user interface for the MRUI quantitation package. MAGMA, 2001. 12(2-3): p. 141-52.
25. Stefan, D., et al., Quantitation of magnetic resonance spectroscopy signals: the jMRUI software package. Measurement Science & Technology, 2009. 20(10).
26. Available from: http://www.jmrui.eu/.
27. Provencher, S.W., Estimation of metabolite concentrations from localized in vivo proton NMR spectra. Magn Reson Med, 1993. 30(6): p. 672-9.
28. von Kienlin, M., et al., 2D-spatial/2D-spectral spectroscopic imaging of intracerebral gliomas in rat brain. Magn Reson Med, 2000. 43(2): p. 211-9.
29. Mareci, T.M. and H.R. Brooker, Essential Considerations for Spectral Localization Using Indirect Gradient Encoding of Spatial Information. J Magn Reson, 1991. 92: p. 229-246.
30. Klomp, D.W. and W.K. Renema, Spectroscopic imaging of the mouse brain. Methods Mol Biol, 2011. 771: p. 337-51.
31. Hiba, B., et al., 2D J-resolved spiral spectroscopic imaging at 7 T: application to mobile lipid mapping in a rat glioma. Magn Reson Med, 2004. 52(3): p. 658-62.
32. Alf, M.F., et al., High-resolution spatial mapping of changes in the neurochemical profile after focal ischemia in mice. NMR Biomed, 2012. 25(2): p. 247-54.
33. Weiss, K., et al., Quantitative in vivo 1H spectroscopic imaging of metabolites in the early postnatal mouse brain at 17.6 T. MAGMA, 2009. 22(1): p. 53-62.