Proton Magnetic Resonance Spectroscopy in Multiple Sclerosis

Acquisition

¹H-MRS can be acquired with spatial information preserved (localized ¹H-MRS), as is most commonly done, or obtain from the whole brain by sacrificing spatial information¹⁴. Point resolved spectroscopy (PRESS)¹⁵ and stimulated echo acquisition mode (STEAM)¹⁶ are the two most commonly used localization techniques in ¹H-MRS. Spectral contamination from extrameningeal tissues is minimized by suppressing the signals from the outer volume¹⁷. Figure 1 demonstrates the placement of eight outer volume suppression bands. The PRESS localization provides better signal-to-noise ratio (SNR) compared to STEAM technique. However, STEAM allows shorterTE- than PRESS.

Localized ¹H-MRS is acquired in either a single voxel or multi-voxel mode. In single voxel MRS, spectral data are acquired from one location at a time. The volume of this region is typically 1 to 8 cm³. This approach involves shorter acquisition and processing times. High-quality spectrum is relatively easy to acquire in this mode. Multi-voxel mode, also referred to as ¹H-MRS imaging (¹H-MRSI) or chemical shift imaging (CSI) allows simultaneous acquisition of ¹H-MRS data from multiple voxels and allows mapping regional distribution of metabolite concentrations and generate metabolic images. Multi-voxel MRS can be acquired in 2D or 3D and generally offers superior spatial resolution (< 1 cm³) compared to single voxel. Generally, ¹H-MRSI is acquired from a volume that is prelocalized using either PRESS or STEAM. It is also possible to acquire ¹H-MRSI without any prior volume or slab localization (see, for example, Sharma et al.¹⁸). ¹H-MRSI involves relatively long acquisition times, which, however, can be considerably shortened by reduced k-space encoding¹⁹. One way of accelerating ¹H-MRSI acquisition is to employ fast spectral-spatial encoding by using echo-planar-based techniques^20,21. With multi-slice and multi echo sequences, it is possible to acquire spectra from whole brain²¹ within reasonable scan times. Combining EPI-based techniques with recent advances in parallel MRI, such as sensitivity encoding (SENSE)²² or generalized autocalibrating partially parallel acquisitions (GRAPPA)²³ with radiofrequency (RF) coil arrays can further accelerate ¹H-MRSI data acquisition^24-26. However, the data acquired with these approaches generally suffers from reduced SNR. Determination of absolute metabolite concentration is also somewhat problematic with parallel acquisition schemes.

¹H-MRS can be acquired at either long or short TE. Long TE ¹H-MRS provides a well-defined baseline by suppressing peaks arising from short T2 metabolites, such as lipids and macromolecules, and makes spectral quantification simpler and more robust. Long TE ¹H-MRS may be preferable if the interest is in the detection of the three major resonances: NAA, Cr, and Cho. Because of their relatively short T2 relaxation times, signals from metabolites, such as Glx, lipids, macromolecules and other myelin breakdown products can be detected only at short TE. Even though the broad spectral baseline in short TE ¹H-MRS makes the spectral quantification more difficult, this approach has the potential to provide important information about altered tissue biochemistry that may not be available at long TE⁴. There is an increasing interest in short TE ¹H-MRSI for visualizing short T2 metabolites, which should result in improved pathological characterization of tissues³. As an example, single voxel spectra acquired at 3 T with short and long TE, with signal mostly localized to white matter of normal brain, are shown in Figure 2.

The macromolecular and lipid resonances exhibit significant overlap with the other metabolites. By nulling the metabolite resonances with an inversion recovery sequence, it is possible to improve the detection of these short TE resonances. This method was employed by Mader et al. ²⁷ for separating lipids from other metabolites in acute MS lesions.

While localized techniques can be used to acquire spectrum from the volume of interest (VOI), non-localized sequence can provide MRS from the entire head. The strong broad lipid signals arising from non-brain tissues obscure the resonances from other metabolites. However, by subtracting the data acquired with and without NAA nulling (through an inversion recovery sequence), it is possible to remove/minimize strong lipid signals and visualize resonances from other metabolites¹⁴. Because NAA is restricted to neurons/axons, it is implicitly localized to brain. However, the presence of other resonances in non-brain tissues, compromises the quantification of metabolites such as Cr. Thus this approach is reliable only when quantifying the whole brain NAA (WBNAA) concentration¹⁴. Due to severe magnetic field inhomogeneities present at air-tissue interfaces this technique captures about 90% of the NAA signal intensity in the brain²⁸. Albeit WBNAA measurements do not provide any information about the spatial distribution of metabolite concentrations, this approach eliminates image registration problems from scan-to-scan in serial studies. The reproducibility of WBNAA levels across institutions and MR scanners and field strengths has been shown to be satisfactory²⁹. WBNAA measurements can be used to quantify the global neuronal and axonal injury to monitor the effect of neuroprotective agents in MS ³⁰.

Quantification

The inherently quantitative nature of MRS allows objective monitoring of disease state and response to therapeutic intervention. However, this requires accurate spectral quantification. Expressing the metabolite levels as a ratio relative to Cr is one of the most common and simplest ways of spectral quantification. However, the interpretation of the metabolite ratios becomes equivocal if the concentration of Cr varies with pathology². In addition, representing brain metabolite concentrations as ratios does not provide complete information and could lead to significant errors^{31, 32}. Therefore absolute quantification of metabolite concentrations is highly desirable for proper interpretation of the MRS data. Nevertheless the NAA/Cr ratio is commonly used in MS ³³. Among the methods for estimating the absolute metabolite concentrations, use of internal water as a standard has been shown to provide acceptable accuracy and inter-laboratory reproducibility^{2, 34}. However, this method depends on knowledge of water distribution and relaxation times in different tissues. Absolute concentrations can also be measured using the LCModel.³⁵ This method analyzes the in vivo spectrum as a linear combination of spectra from individual metabolites contained in a solution (model spectrum). The model spectrum is acquired using the same sequence as that used for in vivo spectra and provides the maximum prior knowledge for the spectral analysis. LCModel uses a nearly model-free constrained regularization method that automatically accounts for the fitted baseline and spectral lineshape to be consistent with the data³⁶.

Usually, the ¹H-MRSI quantification procedures involve extensive computational time, and in some cases, significant human intervention. Currently, the most commonly used quantification methods are based on line fitting using a non linear least square optimization, such as the Levenberg-Marquardt algorithm^{35, 37-39}. The relatively long computational time becomes even more problematic when considering 3D ¹H-MRSI and scans from multi-center clinical trials where a large amount of data needs to be processed. For ¹H-MRSI to become a routine clinical tool, complete automation of quantification procedures and reduction of processing times close to real-time are necessary. The processing time could be reduced by using high performance computing environment. However, this is expensive, involves development of complicated algorithms, and is not an option in routine laboratory/clinical settings. Another approach is to develop fast methods, such as those based on artificial neural networks (ANN). ANN has been used to develop automated methods to quantify ¹H-MRS data^{40, 41}. However, these techniques are limited to either single voxel ¹H-MRS or magnitude spectral data acquired at long TE using lineshapes that may not be completely appropriate⁴². Recently, Bhat et al.⁴³ have employed radial basis function neural networks (RBFNN) to demonstrate the feasibility of quantifying short TE, phased ¹H-MRSI. This method is, however, limited to ratios, and thus does not allow absolute concentrations to be measured.

A number of excellent software packages, both free and commercial, are available for spectral quantification. The most commonly used analysis packages include LCModel ³⁵, jMRUI ⁴⁴, and MIDAS ⁴⁵. Most of these packages involve minimal human intervention and are capable of providing absolute concentrations, but involve long processing times. Different analysis methods and their strengths and weaknesses in computing absolute brain metabolite concentrations were recently reviewed ³¹.

Due to relatively large spectroscopic voxels, spectra have contributions from different types of brain tissues, particularly cerebrospinal fluid (CSF). For accurate estimation of metabolite concentrations, it is important to determine the relative volumes of each tissue that contributes to a given spectroscopic voxel. This requires combining the MRSI data with tissue segmentation, based on the high resolution MRI data³⁷.

Lesions

There is a general consensus that chronic MS lesions exhibit reduced NAA levels, indicating neuronal/axonal loss. This is particularly true in those lesions that also appear hypointense on T1-weighted images (also commonly referred to as “black holes”⁴⁶). In one study, metabolite concentrations from chronic lesions of patients with relapsing-remitting (RR)MS (n=9), secondary progressive (SP)MS (n=10), primary progressive (PP)MS (n=6), and benign MS (BMS) (n=5) were measured and compared with those from white matter of a normal control group (n=9)⁴⁷. A highly significant reduction in NA (the sum of NAA and NAAG) was found in RRMS, PPMS, and SPMS patients compared to controls. However, in patients with BMS there was no significant difference in NA compared to controls. A significant inverse correlation between NA from lesions in MS patients and disability was observed (r=-0.364, 0.05>P>0.02)⁴⁷. Increased Cr and Cho levels were also observed in lesions, suggesting (a) ongoing gliosis and remyelination in isointense lesions on T1-weighted MR images and (b) membrane turnover (de- and remyelination) ⁴⁸.

Acute MS lesions, as detected by gadolinium (Gd) enhancement, demonstrate a range of abnormalities on ¹H-MRS. These abnormalities include reduced NAA, increased Cho, and presence of lipids^{2, 49-52}. Histopathology on biopsied brain tissues demonstrates reduced NAA levels even in the acute phase of the disease⁵³. This could be the result of axonal loss or reflect reversible axonal dysfunction due to functional impairment^{52, 54-56}. Accordingly, serial MRI (with Gd administration) and ¹H-MRSI studies showed transient changes in NAA levels in some acute lesions, which recovered with time². Narayana et al.² have observed that metabolite levels, including NAA, reach their minimum value when lesion volume reaches its maximum. This suggests that the observed decrease in metabolite concentrations, at least partially, can be attributed to resolution of edema within the lesions and is not solely due to compromised tissue metabolism. Cho levels are shown to increase in acute MS lesions^{2, 51, 57}, as well as Lac and mI resonance intensities ^{52, 58}. The presence of lipids in some, but not all lesions was reported in RRMS subjects by Narayana et al.², suggesting the presence of active demyelination/remyelination.

Normal Appearing White Matter

It is now generally agreed, based on MRS^{3, 12}, MT⁵⁹, and DT⁶⁰ studies that the NAWM in patients with MS is abnormal. By exploiting the large spatial coverage allowed by multi-voxel ¹H-MRS, it has been demonstrated that metabolic abnormalities in MS patients are more diffuse and are not restricted to lesion sites alone^{2, 37, 48, 61-64}. ¹H-MRS studies have clearly demonstrated reduced NAA levels in NAWM, suggesting axonal/neuronal loss/dysfunction as the underlying pathologic substrate. The reduced NAA levels on ¹H-MRS have been confirmed by histopathology as secondary to axonal loss^{7, 65}. In serial ¹H-MRS studies, increased Cho and lipid levels were observed from NAWM regions that subsequently went on to develop MRI-visible lesions^{2, 66}. These observations were consistent with MRI studies that showed alterations in MT ratio (MTR) values prior to appearance of macroscopic lesions on MRI⁶⁷. In addition, based on increased mI⁶⁸, Cr and Cho levels, increased cellularity (gliosis, inflammation) was also postulated to occur in NAWM of MS patients⁴⁸.

In a preliminary study, annual MRSI was obtained for two years from 20 patients and 10 healthy controls to characterize the metabolite changes in early RRMS⁶⁹. The concentrations of NAA, Cr, Cho, mI and Glx were estimated in both NAWM and cortical gray matter (cGM). At baseline, the concentration of NAA was 7% lower in NAWM of MS patients compared to the control group and 8.7% lower in cGM. Tissue metabolite profiles did not significantly change in MS subjects nor differed between MS and healthy controls, with the exception of NAA concentrations that tended to recover from baseline. This study suggests that in the early phases of the disease the neuronal/axonal damage may be at least partially reversible. Leary et al.⁷⁰ have performed single voxel ¹H-MRS of NAWM in PPMS (n=24) patients and in 16 age-matched controls. NAA/Cr was significantly lower in NAWM from patients than in the white matter from controls suggesting that axonal loss occurs in NAWM in PPMS patients and may well be a factor associated with disease progression in this disease phenotype. No significant difference was observed in Cr concentration between patients and controls. A multi-center ¹H-MRSI study of 40 PPMS patients showed significantly lower NAA/Cr compared with healthy volunteers, but no significant difference between lesions and NAWM was observed³³. Twenty four of these patients had evident lipid peaks in non-lesion brain tissue locations. Figure 3 shows, as an example, the presence of strong lipid peaks in the spectra from NAWM of a PPMS patient. In another study, metabolite concentrations were obtained with 3D MRS and the VOI centered on the corpus callosum in 11 RRMS patients and nine controls⁷¹. In NAWM, the concentration of NAA was observed to be 9% lower in patients compared to controls. The same study showed an increase in Cr and Cho by 22% and 32%, respectively. These elevated Cho levels differentiated patients from controls with 100% specificity and higher than 90% sensitivity.

Suhy et al.⁷² compared metabolite levels in PPMS (n=15) and RRMS (n=13) patients with those of controls (n=20). Cr in NAWM of PPMS patients was higher compared to controls (P=0.035), and RRMS patients (P=0.038). Also, Cr increased in focal MRI lesions from PPMS compared to lesions from RRMS (P=0.044) and compared to control white matter (P=0.041). NAA was similarly reduced in NAWM from PPMS and RRMS compared to controls. The increased Cr in NAWM in PPMS relative to RRMS is consistent with the notion that the progressive disability in PPMS may be associated with gliosis and axonal loss, whereas disability in RRMS may be secondary to an increased burden of acute inflammatory lesions and axonal loss⁷².

Data on NAWM in clinically isolated syndrome (CIS) is limited. In 14 of 42 patients with a single acute episode of optic neuritis, Tourbah et al.⁷³ reported decreased NAA, increased Cho, and the presence of lipids. In 96 patients with CIS suggestive of MS, Fernando et al.⁷⁴ reported increased concentration of mI relative to controls, a finding which is suggestive for the presence of gliosis. No significant difference in other metabolites was observed. In a prospective study on patients with CIS suggestive of MS, Wattjes et al.^{75, 76} studied 31 patients and 20 controls. These authors observed 8.1% decrease in NAA (P= 0.012) in the NAWM of patients relative to controls. However, changes in Cr, Cho, and mI were not observed.

Gray Matter

Cortical pathology in MS is well recognized^{77, 78}. For a number of reasons GM lesions are difficult to visualize on MRI. Using advanced pulse sequences, such as double inversion recovery ^{79, 80} and phase sensitive inversion recovery sequence, it is possible to display some, but not all, cortical lesions ^{81, 82}. Involvement of GM has been demonstrated by MTR⁸³, DTI^{84, 85}, and ¹H-MRS studies^{18, 68, 86, 87}.

¹H-MRS of cGM is technically challenging because of its proximity to CSF and partial volume averaging from white matter⁸⁸. In addition, many of the MRS studies that employ outer volume suppression bands for minimizing contributions from non-neural tissues make it difficult to acquire MRS from the cGM that is close to outer surface of the brain (see Figure 1). In spite of these limitations, MRS studies of cGM have been reported by a few groups. For example, Sharma et al.¹⁸, by combining a double inversion recovery sequence, in which both CSF and white matter were suppressed, demonstrated abnormalities in the cGM close to the midline fissure in MS patients using MRSI. By combining MRSI with image segmentation, this study has minimized contributions from WM and CSF to the VOI and showed in 52 RRMS patients marked lipid peaks arising from the normal appearing GM, consistent with the presence of active demyelination/remyelination. However, no significant changes in major metabolite concentrations were observed compared to controls, suggesting no significant neuronal loss. The lack of significant changes of metabolites in cGM is consistent with another more recent report⁸⁹.

Using short TE MRSI in 16 RRMS patients with short disease duration (mean disease duration of 1.8 years; range 0.6-2.8 years), Kapellar et al.⁸⁶ reported lower NAA (P = 0.01) and mI (P = 0.04) in the cGM relative to controls. The reduced NAA in cGM suggested mild, but widespread, neuronal dysfunction or loss early on in the course of the disease⁸⁶. In another study from the same group of 24 clinically early (median EDSS (Expanded Disability Status Scale) 1.2; range 0.0-3.0) RRMS patients, Chard et al.⁶⁸ reported significantly reduced NAA, Cho, and Glx. They also observed significant correlations between MS functional composite scores and the metabolite concentrations in cGM: Cr (r(s) = 0.524, P = 0.009), cGM Glx (r(s) = 0.580, P = 0.003). However, no correlation was observed between clinical disability and the level of NAA, suggesting that reduced NAA reflects neuronal dysfunction, rather than neuronal loss early on in RRMS.

GM NAA deficits were found in the progressive form of disease, which, however, did not significantly correlate with EDSS⁹⁰. In another study, absolute concentrations of NAA, Cr, and Cho were measured in the occipital-parietal cortex of RRMS (n=15) and SPMS (n=15) patients and healthy subjects (n=8)⁹¹. Significantly lower values of NAA, Cr and NAA/Cr ratio were detected in SPMS patients than in RRMS and control subjects (P<0.01), suggesting that the pathological process in MS extends into cGM, particularly in the progressive form of the disease. Similar observations were reported in a more recent study on a small MS cohort⁸⁷.

The levels of metabolites in the very early stage of the disease appear to be dependent on lesion activity. In CIS suggestive of MS, Au Duong et al.⁸⁸ reported that the metabolite levels in GM are associated with the presence of enhancing lesions: when absent, there was no statistically significant difference in the GM metabolite levels between patients and controls; when present, a relative decrease in NAA level and an increase in Cho level were observed. This suggests that metabolite alterations seen in the GM in the early phase of the disease may be secondary to the inflammatory process, rather than to ‘primary’ neuronal dysfunction.

MS also affects deep GM structures^{92, 93}. ¹H-MRS studies of the thalami were performed in RRMS (n=14) patients and age-matched healthy controls (n=14). Average NAA concentration in the thalami of RRMS patients was approximately 11% lower relative to controls (p<0.05). In addition, about 25% lower mean normalized thalamic volumes was observed in patients compared to controls (p<0.005). Decreases in thalamic NAA concentrations strongly correlated with thalamic volume loss⁹⁴. The reduction of both NAA concentration and thalamic volume suggests a neurodegenerative component to MS that is increasingly recognized in the disease ⁹⁵.