Various MRS Application Tools for Alzheimer Disease and Mild Cognitive Impairment

SUMMARY: MR spectroscopy is a noninvasive technique that allows the detection of several naturally occurring compounds (metabolites) from well-defined regions of interest within the human brain. Alzheimer disease, a progressive neurodegenerative disorder, is the most common cause of dementia in the elderly. During the past 20 years, multiple studies have been performed on MR spectroscopy in patients with both mild cognitive impairment and Alzheimer disease. Generally, MR spectroscopy studies have found decreased N-acetylaspartate and increased myo-inositol in both patients with mild cognitive impairment and Alzheimer disease, with greater changes in Alzheimer disease than in mild cognitive impairment. This review summarizes the information content of proton brain MR spectroscopy and its related technical aspects, as well as applications of MR spectroscopy to mild cognitive impairment and Alzheimer disease. While MR spectroscopy may have some value in the differential diagnosis of dementias and assessing prognosis, more likely its role in the near future will be predominantly as a tool for monitoring disease response or progression in treatment trials. More work is needed to evaluate the role of MR spectroscopy as a biomarker in Alzheimer disease and its relationship to other imaging modalities.

M R spectroscopy is a noninvasive technique that permits the estimation of the concentrations of various compounds (metabolites) in the human brain in vivo. To be detectable by in vivo MR spectroscopy, a compound must be a relatively small molecule present in at least the low-millimolar concentration range. MR spectroscopy may be performed by using the signals from a number of different nuclei, including phosphorus ( 31 P), carbon ( 13 C), and fluorine ( 19 F), 1 but 1 H-MR spectroscopy (proton MR spectroscopy) has become the most prevalent since the early 1990s because of its higher signal sensitivity, better spatial resolution, and the fact that (unlike other nuclei) no special hardware beyond that found on standard MR imaging scanners is required. 2 During the past 2 decades, numerous studies have used 1 H-MR spectroscopy to study both Alzheimer disease (AD) and mild cognitive impairment (MCI). This article reviews the information content of 1 H-MR brain spectra, gives an update on the current methods used to collect MR spectroscopy (and MR spectroscopic imaging [MRSI]) data, and describes the results of MR spectroscopy research studies in patients with MCI and AD.

Brain Metabolites Detectable by In Vivo MR Spectroscopy
Various metabolites are detectable at 1.5 or 3T with 1 H-MR spectroscopy in normal brain, including the prominent resonances of total NAA (tNAA), trimethylamines (TMA), total creatine (tCr), and signals from myo-inositol (mI), glutamate (Glu), and glutamine. Lactate is not usually seen in normal brain but is detectable in pathologies that cause its concentration to increase. Using special techniques (ie, spectral editing), one can detect other compounds such as ␥-aminobutyric acid (GABA) and glutathione. The following is a description of the metabolites that have been the main interest to date in studies of AD and MCI: tNAA (2.01-ppm singlet), which is estimated to have a healthy adult brain concentration on the order of 10 -12 mmol/L, is usually the largest signal in the spectrum. 3 NAA is an amino acid synthesized in neuronal mitochondria from aspartate and acetyl-coA and is primarily seen in neurons, axons, and dendrites within the central nervous system. 4 Therefore, NAA is commonly referred to as a "neuronal marker," indicative of neuronal density and viability. However, some in vitro studies have shown that NAA can also be found in immature oligodendrocytes and astrocyte progenitor cells. 5 Furthermore, some longitudinal studies suggested that NAA undergoes reversible changes in patients, suggesting it can vary for reasons other than neuronal density. 6,7 Overall, the balance of evidence suggests that NAA is present predominantly in neurons and that it is a reasonably good surrogate marker of neuronal health in many neurologic and psychiatric disorders. The peak detected by MR spectroscopy also contains a small contribution from N-acetyl aspartylglutamate, so for this reason, it is sometimes referred to as total N-acetyl aspartate (tNAA). In addition, in certain rare diseases, it is thought that the tNAA peak may also contain contributions from N-acetyl neuraminic acid (sialic acid) and uridine-diphosphate-N-acetyl-sugars. 8 TMA (3.20-ppm singlet) are a composite peak predominantly containing contributions from glycerophosphorylcholine (a product of the breakdown of membrane phosphatidylcholine) and phosphocholine (a precursor of phosphatidylcholine). There may also be small contributions from free choline, acetylcholine, carnitine, and acetyl-L-carnitine. Since phosphocholine and glycerophosphorylcholine are involved in both the synthetic and degradation pathways of cell membranes, the amplitude of the TMA peak is sometimes thought to be related to membrane turnover. 9 In addition, it has been shown that glial cells have high levels of TMA. 10 tCr (3.01-ppm singlet, 3.9-ppm singlet) consists of the sum of creatine and phosphocreatine, which are energy metabolites. tCr is frequently used as an "internal reference" to quantify other neurochemicals. However, in normal brain, lower levels of tCr are found in white matter than in gray matter, and higher levels of tCr are found in the cerebellum compared with supratentorial regions. Some studies have suggested that tCr levels increase slowly with age. 11 Therefore, the assumption of constant tCr when used as an internal reference should be viewed with caution. mI (3.3-, 3.5-, and 4.0-ppm multiplets) is a sugar alcohol containing contributions from free mI and mI phosphate. 12 mI is readily detectable in short (eg, Յ35 ms) TE spectra; however, it is usually not detectable at long TEs. Because mI has been suggested to be higher in glial cells than in neurons, 13,14 mI is often considered as a "glial cell marker." mI also may act as a marker of a detoxification agent or it may be an osmoregulator or intracellular messenger. 12,13,15 Glu and glutamine (2.1-2.4-and 3.7-ppm multiplets) structures are complex; hence, their peaks are difficult to separate, both from each other and from other compounds. Therefore, they are frequently detected as a composite "Glx" peak. 16 Glu is the major excitatory neurotransmitter in the brain, and glutamine may play a role in detoxification and regulation of its precursor, Glu, within the astrocyte body. 12 It has been suggested that the Glu signal in brain 1 H-MR spectroscopy primarily reflects the intracellular compartment.
GABA (1.9-, 2.3-, and 3.0-ppm multiplets) is the primary inhibitory neurotransmitter in the brain. 17 However, it is difficult to detect GABA by using conventional 1 H-MR spectroscopy because resonances of GABA overlap with other metabolites such as tCr, Glu, tNAA, and macromolecules. In addition, normal brain GABA concentrations are at relatively low levels (1.3-1.9 mmol/ kg wet weight ). 18 For these reasons, most MR spectroscopy studies to date have used spectral-editing methodologies to reliably detect GABA free from contamination from larger overlapping signals.

Methods for In Vivo MR Spectroscopy in the Human Brain
Spatial localization methods for 1 H-MR spectroscopy consist of single-voxel techniques, in which spectral data are acquired from 1 location at a time, or multivoxel techniques. In multivoxel spectroscopy, referred to as MRSI or chemical shift imaging, spectra from multiple regions are acquired simultaneously.
Single-Voxel Techniques. The 2 most commonly used singlevoxel MR spectroscopy techniques are known as the pointresolved spectroscopy sequence (PRESS) 19 and STEAM. 20 Both techniques use 3 mutually orthogonal section-selective pulses and design the pulse sequence to collect only an echo signal from the point (voxel) in space where all 3 sections intersect. In the STEAM sequence, three 90°pulses are used to provide a so-called "stimulated echo," whereas in the PRESS sequence, one 90°and two 180°r efocusing pulses are applied to create a spin-echo. STEAM and PRESS are generally similar but differ in some respects. The biggest difference is that the PRESS should have approximately a factor-of-2 better signal-to-noise ratio than STEAM and is therefore commonly used. However, it is possible to achieve shorter TE values in STEAM than PRESS (which is important for detecting metabolites with short T2 relaxation times), and the 90°sectionselective pulses in STEAM give better voxel profiles compared with the 180°section-selective refocusing pulses used in PRESS. In addition, STEAM may be particularly useful for high-field brain MR spectroscopy. Figure 1 shows both short and long TE spectra recorded using the PRESS and STEAM sequences in 2 patients with AD.
Although widely used, STEAM and PRESS do have some limitations, particularly when used at high magnetic field strengths (eg, 3T or higher). Due to wavelength effects in volume radiofrequency coils 21 or when using inhomogeneous surface coils for excitation, it may be difficult to obtain uniform radiofrequency transmit (B 1 ) fields. This then results in misadjustment of flip angles and variation of the flip angles within the voxel, resulting in suboptimal excitation. The localization by adiabatic-selective refocusing (LASER) sequence 22 or its modified version "semi-LASER," has been applied to solve these problems. 23,24 The LASER sequence is a modification of the PRESS sequence by using adiabatic (usually hyperbolic secant) refocusing pulses. It consists of a non-section-selective adiabatic half-passage 90°pulse for excitation, and 3 pairs of hyperbolic secant refocusing pulses in 3 directions for localization. Compared with STEAM or PRESS, a more uniform excitation profile and decreased chemical shift displacement errors may be achieved in the LASER sequence. 25 However, compared with the PRESS sequence, the radiofrequency power deposition is higher and the minimum TE is longer, due to the large number of radiofrequency pulses. The semi-LASER sequence consists of a nonadiabatic 90°section-selective pulse and 2 pairs of adiabatic hyperbolic secant pulses for refocusing. Semi-LASER has lower radiofrequency power deposition and shorter TEs than LASER, but some insensitivity to B 1 inhomogeneity is lost. 25 The "SPECIAL" consists of a section-selective inversion pulse followed by a spin-echo sequence, with each pulse applied in a different direction. 26 The sequence combines the short TE achievable with STEAM with the full signal intensity provided by PRESS. The sequence collects full-intensity signal from a column defined by the intersection of the selected sections of the 90°and 180°p ulses. 25 The section-selective inversion pulse is applied to every other TR, so that a minimum of 2 scans is required to achieve full spatial localization. 25 The main limitation of spin-echo full-intensity acquired localized spectroscopy is that it is not a "singleshot" localization method and is, therefore, potentially susceptible to head motion or other instabilities, possibly leading to localization errors.
Multiple-Voxel (Spectroscopic Imaging) Techniques. Singlevoxel MR spectroscopy is readily available on nearly all MR imaging scanners, is rapid and relatively easy to perform, and has been used in most AD studies to date. However, single-voxel MR spectroscopy studies are very limited in terms of both coverage and spatial resolution. For instance, single-voxel MR spectroscopy studies of the brain are often limited to 1 or 2 regions and therefore cannot assess the spatial distribution of metabolites. In contrast, MRSI can provide an assessment of the spatial distribution of the various metabolites in addition to their relative concentrations within a voxel.
MRSI can be acquired in 2D 27,28 or 3D and generally offers superior spatial resolution (Ͻ1 cm 3 ) compared with single-voxel spectroscopy. An example of 1 section from a multisection 2D MRSI dataset from a patient with MCI is shown in Fig 2. Generally, 1 H-MRSI is acquired from a volume that is prelocalized by using either PRESS or STEAM; however, as in the example in Fig  2, it is also possible to acquire 1 H-MRSI with only section (or 3D slab) localization. 2D or 3D MRSI with conventional phase-encoding techniques involves long acquisition times, particularly if high spatial resolution and coverage are desired. Therefore, a number of approaches for fast MRSI have been developed, such as echo-planar spectroscopic imaging, spiral MRSI, and sensitivity encoded (SENSE) MRSI.
Echo-planar spectroscopic imaging is one of the fastest acquisition techniques for MRSI. An oscillating read gradient is applied during data readout with each data point in the time-domain corresponding to a bipolar lobe of the read gradient, thus both spectral and spatial information can be collected simultaneously. 29 The oscillating read gradient can be viewed as repeatedly collecting 1 line of k-space at different time points. Conventional phase-encoding is then applied in the other 1 or 2 directions to extend the experiment to either 2 or 3 spatial dimensions, respectively. The echo-planar spectroscopic imaging readout reduces the number of phase-encoding steps by an order of magnitude compared with conventional MRSI, thereby achieving a large scan time reduction. Echo-planar spectroscopic imaging encoding can be implemented with slab-, multisection-, or PRESS-based excitation. 30 As an alternative to echo-planar spectroscopic imaging, spiral-encoded MRSI also applies a field gradient during data acquisition, but unlike echo-planar spectroscopic imaging, it traverses k-space in an in-to-out spiral pattern. 31 Spiral-encoded MRSI has a number of potential advantages compared with the rectilinear k-t sampling pattern of echo-planar spectroscopic imaging, including being less demanding on gradient performance, but to date, it has not been widely adopted for clinical applications.
SENSE-encoded MRSI combines MRSI with reduced phaseencoding, multiple receiver coils, and "parallel" reconstruction techniques (as in parallel MR imaging) 32 to reduce scan time. The  basic principle of the parallel imaging technique is to use the inhomogeneous B 1 fields of multiple, phased-array coils to encode some of the spatial information, thereby allowing fewer conventional phase-encoding steps to be used, hence reducing scan time. In SENSE-encoded MRSI, Fourier transformation of undersampled k-space data leads to aliased spectroscopic images from each receiver channel, which can then be unfolded and reconstructed by using the sensitivity profile of each coil to produce a single spectroscopic image with uniform sensitivity. An attractive feature of sensitivity encoding is that it can be combined with any existing MRSI pulse sequence, so that there are no pulse sequence-related SNR losses, unlike those that can potentially occur with other fast MRSI methods. 33 2D and 3D MRSI involve phase-encoding in multiple directions, so SENSE-encoding can also be performed in 2D or 3D (provided that receive arrays with appropriate geometry are available), leading to large scan time reductions. 34

Spectral Editing Techniques
Although spectral editing can be performed in a number of different ways, currently the Mescher-Garwood (MEGA)-PRESS is the most widely used method, usually for the measurement of GABA at 3T. MEGA-PRESS combines PRESS localization with 2 frequency-selective editing pulses (which can also act to suppress water signal), 35 which are alternately turned "on" and "off" in an interleaved fashion throughout the acquisition. For the detection of GABA, usually the frequency-selective editing pulse is placed at 1.9 ppm in the "on" (even) acquisition, and the "off" acquisition actually applies the pulse symmetrically on the opposite side of the water peak (4.7 ppm) at 7.5 ppm. Because the GABA resonance at 1.9 ppm is coupled to the 3.0-ppm resonance, the resulting edited spectrum (derived from the difference between the "on" and "off" spectra) shows a peak at 3.0 ppm, while overlapping resonances are suppressed.
Other, less frequently used "editing" techniques for coupled molecules, include the 2D J-PRESS method, 36,37 chemical shift selective filter, 38 constant time PRESS, 39 TE-averaged PRESS, 40 and maximum-echo sampling methods. 37 Some of these methods may also be combined with MRSI, for instance a 2D MRSI TEaveraged PRESS technique was used to generate Glu maps. 41 Recently, with the advent of techniques for speeding up MRSI acquisitions such as compressed sensing, it has become possible to perform multidimensional MRSI scans (2 spectral, 3 spatial), which have promise for mapping compounds that are difficult to observe with conventional MRSI techniques. 42 Currently, nearly all the techniques mentioned above, other than STEAM, PRESS, and conventional MRSI, are not commercially available and therefore are only performed by research groups with access to these methodologies.

High-Field MR Spectroscopy
Spectral SNR and chemical shift resolution increase with increasing magnetic field strength (B 0 ), though the SNR increase may sometimes be less than the linear improvement predicted by theory. 43 This outcome is most likely explained by the increase in spectral line widths with increasing B 0 . While field homogeneity is usually measured from voxel water line widths, typical metabolite (eg, for Cr or NAA) line widths in the human brain are 3.5 Hz at 1.5T, 5.5 Hz at 4T, and 9.5 Hz at 7T. 44 The metabolite line width also depends on B 0 field homogeneity and metabolite T 2 relaxation time. It has been found that metabolite T 2 relaxation times measured in vivo decrease with increasing B 0 . 45 For instance, the T 2 of NAA drops from approximately 300 -450 ms at 1.5T 43,46,47 to 210 -300 ms at 3T, 43,48 to 185-230 ms at 4T, 45 and 140 ms at 7T. 44 Therefore, high-field MR spectroscopy is best performed at short TEs (such as Յ35 ms). Despite increasing line width, improved SNR and chemical resolution of spectra are shown at 7T compared with 3T, and 3T compared with 1.5T. 43,44 Moreover, resonances from coupled spin systems such as GABA, Glu, and glutamine are better demonstrated at high field. Therefore, performing MR spectroscopy studies at the highest field available is recommended.

Spectroscopic Applications in MCI and AD
AD is the most common cause of dementia; MCI is the transitional state between normal aging and AD. The articles discussed below were the result of a PubMed search by using the phrases "Alzheimer disease or mild cognitive impairment" and "MR spectroscopy." The search was restricted to the English language (181 results). The articles discussed are classified into the following categories: diagnosis, differential diagnosis, therapy, and other applications.

Diagnosis
The Table shows methodologic details and results of studies that compared AD, MCI, and healthy controls. Reduction of NAA levels compared with age-matched healthy controls was the most frequent 1 H-MRS finding in AD. 49,50 Reduced NAA/Cr ratios have been shown in the posterior cingulate gyrus, mesial temporal lobe, occipital lobe, parietal lobe, and frontal lobe. [51][52][53][54][55][56][57][58][59][60][61][62] The "absolute" concentration of NAA was also decreased in the parietal lobe, occipital lobe, mesial temporal lobe, frontoparietal region, and hippocampus. [63][64][65][66][67][68][69] These results are consistent with in vitro studies performed on postmortem AD brains, which showed that NAA decreases correlated with the severity of neuropathologic findings, such as amyloid plaques, neurofibrillary tangles, and the presence of apolipoprotein E genetic markers. 60 Therefore, reduction of NAA levels might reflect either a loss of the neuronal cells or neuronal function, or both.
Increases of mI have also been reported in several anatomic locations in AD, indicative of increased glial cell content. Increased mI has been reported most often in the posterior cingulate gyrus, temporal-parietal area, parietal white matter and occipital lobes. 51,60,61,65,70 However, 2 studies reported no significant differences in mI levels between patients with AD and healthy controls. 55,71 Some researchers have used the ratio of NAA/mI to increase the sensitivity of 1 H-MR spectroscopy to metabolite changes in AD. 61,72,73 The NAA/mI ratio has been shown to be the more accurate MR spectroscopic measurement to differentiate patients with AD from healthy elderly.
There are conflicting reports on the TMA level in patients with AD. Some studies showed elevated TMA levels in AD, 61,74-76 while others report decreased TMA levels 66,69,77 or no change. 53,60,62 It has been suggested that these disparate findings may be the result of possible apolipoprotein E allele effects on membrane metabolism or breakdown, differences in MR spectroscopy methods (in particular TE), or variations in anatomic voxel placement. 68 While the tCr peak is generally thought to be stable in AD, some studies have demonstrated decreased tCr levels in patients with AD versus healthy controls in the occipital lobe 69 and increased tCr levels in the parietal and occipital lobes. 65 The Glx peak has only been investigated in a few studies; these have mostly reported reduced Glx levels in patients with AD compared with controls in the posterior cingulate gyrus and parietooccipital white matter, 58 though 1 study reported no difference in Glx between patients with AD and healthy controls. 61 Finally, 1 study demonstrated decreased levels of glutathione in the right frontal cortex of female patients with AD compared with healthy female controls and decreased glutathione levels in the left frontal cortex of male patients with AD. 78 In this study, the glutathione level also showed a trend toward reduction in patients with MCI compared with healthy subjects, though the difference was not statistically significant. 78 Generally, metabolic differences between patients with AD, MCI, and controls are fairly small, and appreciable scatter (overlap) exists between groups. Therefore, MR spectroscopy in isolation usually cannot be used in individual subjects for the diagnosis of AD. Metabolite concentrations should be corrected for CSF contamination because brain atrophy is typically very significant in both the elderly healthy control population and particularly in patients with AD.

Differential Diagnosis
A number of studies have compared subcortical ischemic vascular dementia with AD. mI/tCr was found to be higher, and tNAA/tCr, lower, in patients with AD compared with those with subcortical ischemic vascular dementia. 79,80 It has been reported that there is a significant correlation between Mini-Mental State Examination score and tNAA/mI and tNAA/tCr in patients with AD but that patients with subcortical ischemic vascular dementia showed no correlations. 81 In 1 study of AD and frontotemporal dementia, tNAA/tCr was reduced in the posterior cingulate gyrus in both patients with AD and frontotemporal dementia; however, the patients with AD showed a posterior dominant decrease, whereas there was a frontal predominant decrease in the patients with frontotemporal dementia, 82 while another study found no significant metabolic differences between AD and frontotemporal dementia in the posterior cingulate gyrus. 79 However, it was found that tNAA/tCR was higher in patients with dementia with Lewy bodies than AD in the posterior cingulate gyrus. 79 Finally, 1 study found that MR spectroscopy can differentiate between AD and MCI, with the strongest effect seen with the tNAA/tCr ratio in the

Effects of Therapy
A number of studies have investigated the effect of the acetylcholinesterase inhibitor donepezil in AD. [84][85][86][87] It was found that tNAA and tNAA/tCr tended to be higher in the donepezil-treated patients compared with a placebo, 86 although 1 study found that tNAA, TMA, tNAA/tCr, TMA/tCr, and mI/tCr were all decreased in patients with AD after treatment. 84 An MRSI study of the muscarinic acetylcholine receptor agonist xanomeline found no metabolic differences before or after treatment in either study drug or placebo groups but did find a positive correlation between parietal lobe gray matter TMA/tCr ratio and cognitive performance. 88 Finally, in a small trial (n ϭ 10) of the alkaloid galantamine, levels of Glu and Glu/tCr in the right hippocampus increased after 4 months of treatment, and these changes were associated with increased cognitive performance. 89

Other Applications
One study of MCI and patients with early AD found correlations between various cognitive measures (eg, verbal learning performance, memory) and cortical metabolite ratios. 90 Another compared MR spectroscopy and diffusion tensor imaging findings and found a positive correlation between mI/tCr and fractional anisotropy values and a negative correlation between tNAA/tCr and mean diffusivity values in patients with AD. 91 Finally, in a study of 15 control subjects and 22 patients with AD, it was reported that correlations exist between elevated basal ganglia TMA signal and poorer performance on learning tasks, with more significant findings in male patients with AD, suggesting sex-specific effects. 92

CONCLUSIONS
Although an abundance of studies have shown metabolic changes in the brain in subjects with MCI and AD, at present, MR spectroscopy is little used in the clinical evaluation of subjects with dementia. This is probably due to several reasons, including lack of standardized methodology, overlap of spectral patterns between different pathologies (ie, relative lack of specificity), lack of reimbursement, and lack of treatment options in most dementias.
Arguably the most important role for MR spectroscopy may be in predicting the development of dementia in patients with MCI, information that is highly valued by patients and their families. The relative value of MR spectroscopy, performed with the best available contemporary methods, needs further evaluation in this regard; its performance should be compared with other imaging measures from MR imaging and positron-emission tomography, in particular amyloid imaging methods, as well as other biomarkers and clinical testing. For the time being, MR spectroscopy may find a greater role as a surrogate marker of brain metabolic health in clinical trials, perhaps both for patient selection and monitoring outcome.