OUP user menu

Longitudinal quantitative proton magnetic resonance spectroscopy of the hippocampus in Alzheimer’s disease

Ruth M. Dixon, Kevin M. Bradley, Marc M. Budge, Peter Styles, A. David Smith
DOI: http://dx.doi.org/10.1093/brain/awf226 2332-2341 First published online: 1 October 2002


Changes in metabolites detected by proton magnetic resonance spectroscopy (1H MRS) of the brain have been demonstrated in Alzheimer’s disease. Our objectives were, first, longitudinally to measure absolute concentrations of metabolites in both hippocampi, the sites of early Alzheimer’s disease, in patients with clinical Alzheimer’s disease and controls; secondly, to separate the relative contribution of atrophy and metabolite concentration change to overall signal change; and, thirdly, to determine whether metabolite concentrations in the hippocampus relate to cognitive scores. 1H MR spectra were acquired from a single voxel (12 × 15 × 25 mm3 = 4.5 ml) aligned to the long axis of each hippocampus in nine probable or possible Alzheimer’s disease subjects diagnosed according to the National Institute of Neurologic and Cognitive Disorders and Stroke (NINCDS) compared with 14 age‐matched NINCDS‐negative Alzheimer’s disease controls. Metabolite concentrations were corrected for the amount of CSF present in the voxel. Hippocampal volumes were measured at the same time. The same protocol was repeated approximately 1 year later. We found that atrophy‐ corrected hippocampal N‐acetylaspartate (NAA) concentration was lower in cognitively impaired subjects compared with controls. This was significant for the left hippocampus (baseline 87% of control, P = 0.013; and at 1 year 76% of control, P = 0.020). Hippocampal volumes also differed significantly between the groups, and decreased significantly over 1 year in the Alzheimer’s disease group (12%, P = 0.017). The decrease in [NAA] over 1 year was not significant in either group. Discriminant analysis revealed that the best classification of subjects was by including both left NAA concentration and left hippocampal volume. myo‐Inositol signals from these small voxels had poor signal‐to‐noise and demonstrated no significant changes. We conclude that 1HMRS‐detectable metabolites can be quantified from the hippocampi of cognitively impaired individuals, and that hippocampal [NAA] is significantly reduced in Alzheimer’s disease, in excess of atrophy. In our cohort, the differences were more significant for the left hippocampus.

  • Keywords: proton MRS; neurodegeneration; dementia; medial temporal lobe
  • Abbreviations: CAMCOG = Cambridge cognitive examination; Cho = choline; Cr = creatine; 1H MRS = proton magnetic resonance spectroscopy; MI = myo‐inositol; MMSE = Mini‐Mental State Examination; NAA = N‐acetylaspartate; NINCDS = National Institute of Neurologic and Cognitive Disorders and Stroke

Received September 12, 2001. Revised February 27, 2002. Second revision May 6, 2002. Accepted May 7, 2002


There are numerous reports of metabolite changes measured by proton magnetic resonance spectroscopy (1H MRS) in clinically diagnosed Alzheimer cases (reviewed recently by Valenzuela and Sachdev, 2001). A pattern has emerged of decreased N‐acetylaspartate (NAA) and increased myo‐inositol (MI) concentrations. These changes have been observed in most brain regions, usually from relatively large volumes of interest, typically 8–27 ml, in order to obtain good spectra. Since the medial temporal lobe, in particular the hippocampus, usually demonstrates marked pathological changes in Alzheimer’s disease (Braak et al., 1993; Braak and Braak, 1997; Delacourte et al., 1999), this is likely to be the site of greatest metabolic change. Indeed, there is believed to be extensive hippocampal pathology leading to early atrophy of this structure in the pre‐symptomatic phase of Alzheimer’s disease (Jack et al., 1997; Nagy et al., 1999). With the emergence of potential therapeutic agents, a metabolic marker of this early stage of Alzheimer’s disease would have a vital role in monitoring drug intervention.

Most previous reports of MRS of the medial temporal lobe in Alzheimer’s disease calculated metabolic ratios (e.g. Jessen et al., 2000, 2001). However, without absolute measurement of the concentration of metabolites, the relative contribution of atrophy is not clear. Schuff et al. (1997) and Huang et al. (2001) measured absolute metabolite concentrations in the medial temporal lobe and/or the hippocampus. The study of Schuff et al. (1997) showed that the combination of hippocampal volume and hippocampal [NAA] gave better discrimination of Alzheimer’s disease than did either measurement alone. It is known that with atrophy, the ratio of neurones to glial cells changes (e.g. Scott et al., 1992; Leuba and Kraftsik, 1994). NAA is found within neurones (Moffett et al., 1991; Simmons et al., 1991). It has not been found in astrocytes but has been detected in oligodendrocytes in culture (Bhakoo and Pearce, 2000). Interpreting metabolic changes from ratios compared with creatine (Cr) makes the assumption that Cr concentrations are equal in glia and neurones, which is not true, at least for cells in culture (Urenjak et al., 1993).

We studied the changes in NAA (and other metabolites) in the hippocampi of patients with National Institute of Neurologic and Cognitive Disorders and Stroke (NINCDS) criteria (McKhann et al., 1984) for probable and possible Alzheimer’s disease compared with NINCDS‐negative controls over 1 year. Two other longitudinal MRS studies of Alzheimer’s disease patients have been reported (Adalsteinsson et al., 2000; Jessen et al., 2001). Adalsteinsson et al. (2000) reported an overall decline in NAA in cerebral grey matter, which was proportionately greater than the overall grey matter atrophy. The medial temporal lobes were not included in the analysis. Jessen et al. (2001) reported a decrease of NAA/Cr in the medial temporal lobe of Alzheimer’s disease patients.

Kantarci and co‐workers used 1H MRS to measure metabolite ratios from the posterior cingulate, left superior temporal lobe and the medial occipital lobe in groups of controls, probable Alzheimer’s disease and mild cognitive impairment (MCI) (Kantarci et al., 2000). They demonstrated reduced NAA/Cr and increased MI/Cr in the superior temporal lobe and posterior cingulate of Alzheimer’s disease subjects, but the only significant finding comparing the MCI group with controls was an elevation of MI/Cr of the posterior cingulate. These findings are in keeping with the pattern of development of Alzheimer’s disease pathology (Braak et al., 1993). As the MCI cohort was likely to contain a significant proportion of ‘pre‐clinical’ Alzheimer’s disease, the findings emphasize the importance of collecting spectra from the sites of earliest pathological involvement. Since 1H MRS of the entorhinal cortex is not feasible, then the next most promising target for 1H MRS in Alzheimer’s disease is the hippocampus.

The hippocampus is technically challenging to study by MRS, due to its small size and its proximity to areas of air and bone within the cranium, which lead to spectra with low resolution and poor signal‐to‐noise or, alternatively, severe partial volume effects from surrounding tissue. Nevertheless, we were able to obtain interpretable MR spectra from a 4.5 ml voxel slanted along the long axis of the hippocampus from all subjects.

The amount of metabolite in the voxel depends both on tissue concentration and tissue volume, and so we compensated for variable amounts of CSF (resulting from tissue atrophy) by analysing the separate T2 components of water in the voxel and normalizing to the tissue component. The aim of the present study was to determine the absolute concentration of tissue NAA (corrected for atrophy) in the hippocampus of Alzheimer’s disease subjects compared with age‐matched controls, and to repeat the protocol in the same subjects after 1 year. We used the method of Ross and co‐workers (Ernst et al., 1993; Kreis et al., 1993) to obtain absolute concentrations of NAA, choline (Cho) and Cr corrected for tissue atrophy. By combining these concentration measurements and hippocampal volumes measured from MRIs, estimates of total amounts of hippocampal NAA, Cho and Cr could be derived. Ratios of metabolites were also calculated. We compared the MRS‐derived parameters with cognitive scores and with hippocampal volume measurements.



Nine subjects with cognitive impairment (four female, five male, median age 70, range 60–89 years) and 14 age‐matched controls (eight female, six male, median age 74, range 60–83 years) were recruited. All subjects were volunteers to the Oxford Project To Investigate Memory and Ageing (OPTIMA), and will eventually have post‐mortem histology. OPTIMA is a prospective, longitudinal, clinicopathological study of ageing, in both ‘control’ elderly and memory‐impaired subjects. Informed consent was obtained from subjects without cognitive deficit and was given by a close relative in those with a significant deficit. Ethical approval was obtained from both the Central Oxford Research and the Psychiatric Sector Research Ethics Committees.

All subjects performed the Mini‐Mental State Examination (MMSE; Folstein et al., 1975) and The Cambridge Cognitive Examination (CAMCOG; Roth et al., 1988) to assess their cognitive function. Their apolipoprotein E (Apo E) genotype was assessed using standard methodology. The Alzheimer’s disease status of each subject was assessed by the NINCDS criteria (McKhann et al., 1984). Demographic details of Alzheimer’s disease subjects and controls are shown in Table 1 and at the top of Table 2. As hippocampal atrophy and altered [NAA] are not specific to Alzheimer’s disease, those included in the Alzheimer’s disease and control groups had no evidence of potential confounders such as cerebral infarction, depression, fronto‐temporal dementia or substance abuse. The diagnosis of possible Alzheimer’s disease was made for four patients on the basis of having a defect in only one cognitive domain (amnestic syndrome). All Alzheimer’s disease subjects (possible and probable) had Hachinski Ischaemia Scores (Hachinski et al., 1975) of 4 or below. Subjects with vascular risk factors, however, were not excluded from patient or control groups, e.g. they were not excluded on the basis of being smokers or the presence of white matter changes on MRI, since we wished to study a typical cross‐section of this age group.

View this table:
Table 1

Clinical and demographic details of the Alzheimer’s disease patients

SubjectSexApoE genotypeNINCDS statusFirst MRS studySecond MRS study
1 F3/4Possible+782176801976
3 F3/3Probable661767681139
4 M3/4Probable811960821449
5 F4/4Possible602487612381
6 M3/3Probable671785672077
7 M3/3Possible722993‐r‐

‐r‐ = refused second study. +Subsequently progressed to probable Alzheimer’s disease.

View this table:
Table 2.

Results from the two MRS studies, 1 year apart

Normal elderly subjectsAlzheimer’s disease (NINCDS probable + possible) subjects
First studySecond studyFirst studySecond study
n 14 (8F)9 (4F)8 (4F)
Age at 1st study: median (range)74 (60–83)70 (60–89)
MMSE: median (range) (max 30)30 (28–30)30 (27–30)21.5 (5–26)**18.5 (8–24)**
CAMCOG: median (range) (max 107)102 (100–106)103 (96–107)83.5 (28–94)**73.5 (25–82)**,+
Right hippocampus
Volume (ml)3.81 ± 0.653.83 ± 0.453.09 ± 0.76*3.07 ± 0.82*
[NAA] (corrected for atrophy) (mM)9.8 ± 1.79.4 ± 1.29.1 ± 1.68.7 ± 1.3
[Cr] (corrected for atrophy) (mM)9.7 ± 2.110.1 ± 2.010.8 ± 1.910.1 ± 3.4
[Cho] (corrected for atrophy) (mM)2.4 ± 0.62.1 ± 0.42.4 ± 0.51.8 ± 0.3
Total NAA (µmol)37.3 ± 9.535.7 ± 5.329.0 ± 10.826.5 ± 7.4**
Total Cr (µmol)37.1 ± 11.638.2 ± 7.932.9 ± 11.531.8 ± 16.9
Total Cho (µmol)9.3 ± 2.98.0 ± 1.97.3 ± 2.45.4 ± 1.6*
NAA/(Cho + Cr)0.65 ± 0.140.68 ± 0.150.55 ± 0.120.64 ± 0.15
MI/(Cho + Cr)0.16 ± 0.080.22 ± 0.130.20 ± 0.100.17 ± 0.12
MI/NAA0.25 ± 0.120.33 ± 0.230.37 ± 0.170.27 ± 0.18
T2 ‘brain’ water (ms) 82 ± 484 ± 682 ± 885 ± 14
Left hippocampus
Volume (ml)3.79 ± 0.733.72 ± 0.432.92 ± 0.63*2.44 ± 0.54**,+
[NAA] mM (corrected for atrophy)9.8 ± 1.29.7 ± 1.98.6 ± 0.7*7.4 ± 2.1*
[Cr] mM (corrected for atrophy)10.5 ± 2.49.5 ± 2.69.2 ± 3.611.0 ± 2.6
[Cho] mM (corrected for atrophy)2.6 ± 0.62.3 ± 0.52.3 ± 0.72.8 ± 0.9
Total NAA (µmol)37.0 ± 7.036.5 ± 9.925.1 ± 6.1**17.8 ± 6.0**
Total Cr (µmol)39.6 ± 11.335.1 ± 9.127.4 ± 12.826.5 ± 3.1**
Total Cho (µmol)10.1 ± 3.28.5 ± 2.06.8 ± 3.4*7.1 ± 2.9
NAA/(Cho + Cr)0.60 ± 0.100.68 ± 0.160.65 ± 0.180.46 ± 0.16**
MI/(Cho + Cr)0.16 ± 0.100.22 ± 0.140.23 ± 0.150.20 ± 0.13
MI/NAA0.29 ± 0.210.34 ± 0.240.45 ± 0.350.44 ± 0.29
T2 ‘brain’ water (ms) 85 ± 485 ± 690 ± 1893 ± 10*

Data are given as mean ± SD, unless otherwise indicated. NAA = N‐acetylaspartate; Cho = choline, Cr = creatine; T2 = transverse relaxation time. **P < 0.01 (Mann–Whitney U test) relative to controls, same study; *P < 0.05 (Mann–Whitney U test) relative to controls, same study; +P < 0.02 (Wilcoxon signed‐ranks test) first versus second study.

Proton MRI and spectroscopy

1H MRS of the left and right hippocampi was carried out for each subject. Imaging and spectroscopy were performed on a 2‐T magnet (Oxford Magnet Technology, Eynsham, Oxon, UK) interfaced to an Avance Spectrometer (Bruker Medical, Ettlingen, Germany) and a purpose‐built quadrature head coil. Sagittal viewfinder images were obtained, followed by ‘tilted coronal’ images slanted so that the slices were perpendicular to the long axis of the body of the hippocampus (Fig. 1). This utilized a fast spin‐echo sequence [repetition time (TR) 3000 ms, echo time (TE) 13 ms]. Ten slices were obtained with slice thickness of 5 mm and slice separation of 5.5 mm. A T1‐weighted image was obtained with the same positioning and slice parameters [TR 500 ms, TE 10 ms, pulse angle 70°]. A slanted rectangular volume 12 × 15 × 25 mm3 (4.5 ml) was selected so that it was aligned along the long axis of the hippocampus, starting just posterior to the amygdala. This ensured that the majority of the tissue within the voxel was hippocampal. Localized proton spectra were obtained from this voxel with the stimulated‐echo acquisition mode (STEAM) [TR 1.5 s, TE 90 ms, mixing time (TM) = 17 ms, number of averages = 256]. Water suppression was by chemical shift selective pulses during the relaxation delay and in the TM period. The proportions of CSF and brain tissue water in the volume of interest were assessed by collecting single scan, non‐water‐suppressed spectra with a range of TE values from 30 to 1040 ms (TR 5 s). The peak heights of these spectra were fitted to a decaying double exponential in which the longest rate constant was fixed at 0.5/s, i.e. the T2 relaxation rate of CSF (measured in control subjects). The resulting double exponential allowed the proportions of brain tissue water and CSF to be determined, since the relaxation rate of brain tissue water (intracellular + extracellular) was much faster than that of CSF (Ernst et al., 1993; Kreis et al., 1993). A 16‐scan spectrum of water was also collected at TE = 30 ms for accurate quantification (TR 5 s). The left side hippocampus was studied first, followed by the right side. The whole study was completed in ∼45–50 min.

Fig. 1 (A and B) Viewfinder images showing positioning of hippocampal voxels. (C) 1HMR spectrum of the left hippocampus of a subject with probable Alzheimer’s disease. Acquisition parameters are as described in the Methods section.

The relatively long TE of 90 ms was chosen to optimize a number of competing criteria. Water suppression was better at longer TE, which improved the quantification of metabolites. However, T2 relaxation and (in the case of MI) scalar coupling reduce the intensity of signals during the TE. Simulation of the effect of TE on the spectrum of MI showed that a partial refocusing takes place at a TE of 90 ms. This was confirmed by phantom studies (data not shown), and therefore TE = 90 ms was chosen for the study.

Data analysis

Spectra were processed with 5 Hz exponential line broadening, Fourier transformation, manual baseline correction and automated line fitting (WinNMR, Bruker‐Franzen Analytik, Bremen, Germany) to the signals arising from NAA, Cho, Cr and MI. The concentration of each metabolite was calculated by relating its signal area to the signal arising from brain tissue water in the selected volume, corrected for T1 (longitudinal) and T2 (transverse) relaxation of water. Corrections were also made for incomplete T1 relaxation and T2 decay of metabolites, using the relaxation times measured in the hippocampus of normal subjects at the same field strength (Choi and Frahm, 1999). By separately quantifying the brain tissue water and the CSF in the hippocampal voxel, and using the brain tissue water as an internal standard, the metabolite concentration in the volume of interest was corrected for hippocampal atrophy. Quoted ratios of metabolites are uncorrected for relaxation effects or number of contributing protons.

Hippocampal volume measurement

Hippocampal volumes were measured from the T1‐weighted coronal oblique images acquired perpendicular to the long axis of the body of the hippocampus. Manual outlining of the hippocampus was performed as described by Jack et al. (1989) with the modification of Watson et al. (1992) defining the posterior limit as the first image where the crus of the fornix is seen separately from the tail of the hippocampus. Reference was made to the pathological and MRI atlas of Duvernoy (1998). All of the hippocampal volumes were measured by a radiologist experienced in hippocampal volumetry, who was blind to all clinical information.


MRS and hippocampal volume reproducibility were assessed by interspersing 11 studies from a younger healthy control subject (age 35 years) in the set of experimental studies. These measurements were obtained by the same protocol over the 20‐month course of the study (median interval 45 days). They were assessed blindly by the same protocol as the older subjects’ studies. The right and left hippocampal volumes from this subject had coefficients of variation of 9.7 and 5.0%, respectively. The metabolite concentrations from the same subject showed coefficients of variation of 11% (right side) and 11.5% (left side) for [NAA], while the coefficients of variation of [Cho] and [Cr] were ∼20%. Metabolite ratios had coefficients of variation of 20% [NAA/(Cho + Cr)] and 29–44% [MI/NAA and MI/(Cho + Cr)]. We are unable to provide unequivocal evidence that the reproducibility measurements are valid for the older subjects; however, there was no systematic difference in other ‘quality control’ measures, e.g. water proton linewidth, overall signal‐to‐noise or image quality, between the older and younger subjects.

Longitudinal study

All 14 control subjects, and eight of the nine Alzheimer’s disease subjects agreed to return for a second MRS study. The second study took place about a year after the first, with a mean interval for controls of 13.0 months (range 11–15 months) and for subjects of 12.7 months (range 9–18 months). Exactly the same protocol was followed as in the first study. Cognitive testing was also performed close to the date of the MRS study.

Statistical analysis

Statistical differences were assessed non‐parametrically by the Mann–Whitney U test, or the Wilcoxon signed‐rank test (for paired comparisons) and correlations by non‐parametric Spearman’s rank correlation test (SPSS for Windows, version 10.0.0; SPSS Inc., Chicago, IL, USA).


Demographic and clinical details of the Alzheimer’s disease subjects are shown in Table 1. It is clear from the cognitive scores that three subjects were very early cases, since their MMSE and CAMCOG scores were close to, or even above, the accepted test thresholds for dementia. However, all of the nine cases were assessed as probable or possible Alzheimer’s disease by NINCDS criteria.

MR spectra

Typical MR spectra are shown in Fig. 1, together with images indicating the spectroscopic voxels. It should be noted that MI could not be quantified in some of the spectra (five out of 56 control spectra, and four out 35 Alzheimer’s disease spectra), and these cases were not included in the analysis of the MI ratios. In these cases, the MI signal was superimposed on a rolling baseline close to the suppressed water signal, and could not be fitted accurately. In one spectrum, the Cho and Cr signals also suffered from this problem, and were excluded from the analysis, and, in two further spectra (from different Alzheimer’s disease patients), Cho and Cr could not be fitted separately due to poor spectral resolution and so only the sum of (Cho + Cr) was measured. One subject was unable to tolerate the entire study, so only his left side hippocampus was measured on the second occasion. MRS results are shown in Table 2.

First MRS study

In the first study, a significantly lower left hippocampal [NAA] (after correction for atrophy) was found in the Alzheimer’s disease subjects compared with the controls (87% of control, P = 0.013) (Fig. 2, Table 2). The hippocampal volumes were found to be smaller in Alzheimer’s disease subjects (right side, 81% of control, P =  0.033; left side, 77% of control, P = 0.011). The total amount of hippocampal NAA was also significantly lower on the left side in the subjects with cognitive impairment (68% of control, P < 0.0005), and the left hippocampal Cho was also lower (P = 0.024), but these are a combination of the effects of atrophy (smaller hippocampal volumes) and lower tissue metabolite concentrations. Considering metabolite ratios, only the right side NAA/Cr was lower in Alzheimer’s disease subjects (76% of control, P = 0.023). The water T2 components were analysed to reveal that the Alzheimer’s disease cohort had more CSF in the left hippocampal voxel [left, 15 ± 8% compared with 7 ± 5% in controls (P = 0.009); right, 14 ± 9%, compared with 8 ± 7% in controls (P = NS)].

Fig. 2 Filled bars, control subjects, study 1; dotted bars, control subjects, study 2; grey bars, Alzheimer’s disease subjects, study 1; open bars, Alzheimer’s disease subjects, study 2. *P < 0.05; **P < 0.01, compared with controls in the same study.

Second MRS study

The overall findings from the second MRS study were similar to those of the first (Table 2) and, in general, more marked abnormalities were seen. Unfortunately, one Alzheimer’s disease subject declined to return for the second study, and a further subject tolerated only the first half of the study (left side). Despite this reduction in subject numbers, we again found statistically significant differences between the Alzheimer’s disease and control groups in left hippocampal [NAA] corrected for atrophy (76% of control, P = 0.020) and in total hippocampal NAA (right side 74% of control, P = 0.007, left side 49% of control, P < 0.0005). The hippocampal volumes were lower in Alzheimer’s disease subjects (right side 80% of control, P = 0.042; left side 66% of control, P < 0.0005). In the second study, left side NAA/(Cho + Cr) was lower in Alzheimer’s disease subjects (68% of control, P = 0.010), and NAA/Cr was also reduced (68% of control, P = 0.016). No differences were found in the ratios involving MI. The hippocampal voxels of the Alzheimer’s disease patients again contained more CSF than did those of the controls (left, 23 ± 12% Alzheimer’s disease, 7 ± 5% controls, P = 0.001; right, 25 ± 14% Alzheimer’s disease, 7 ± 5% controls, P = 0.002).

Longitudinal changes

The subjects were studied on two occasions about a year apart [median interval (range): controls 13 (11–15) months, Alzheimer’s disease 12 (9–18) months]. Comparing the first with the second study, using the non‐parametric Wilcoxon signed‐rank test, no significant changes were seen in the control group. The only significant change in MR‐measured parameters in the Alzheimer’s disease subject group was a decrease in left hippocampal volume by an average of 0.36 ml (12%, P = 0.017). There was a trend towards a lower total left hippocampal NAA in Alzheimer’s disease subjects, but this did not reach statistical significance (six out of eight subjects showed a decrease, mean decrease 6 µmol, 24%, P = 0.07). Total left hippocampal NAA decreased in seven of the 14 control subjects (P = NS). CAMCOG scores decreased significantly in the Alzheimer’s disease subjects (median change –9.5, P = 0.018), although MMSE showed no significant change (median change –2.0, P = 0.21). For all of the MR‐determined parameters, the changes were greater for the left hippocampus than the right (Fig. 2).

Apo E genotype

Apo E genotype was available for all of the subjects. Subjects with genotypes known to confer increased risk of Alzheimer’s disease (ϵ3/4 or ϵ4/4) were grouped as ϵ4 carriers. Five out of 14 of the control subjects had genotype ϵ3/4; the remainder had ϵ2/3 or ϵ3/3. Four out of nine Alzheimer’s disease subjects had genotype ϵ3/4, one had ϵ4/4 and the remainder were ϵ3/3. No statistical differences were identified in the MRS parameters when the subjects were grouped on the basis of ApoE 4 genotype (ϵ4 carriers versus non‐carriers), although the small number of subjects in the study may mask actual differences.

Correlations between cognitive and clinical scores and MRS‐derived parameters

Given that the differences between control and Alzheimer’s disease individuals were greater for the left hippocampus, only left side results are considered here. Hippocampal volume and total hippocampal NAA were significantly correlated with CAMCOG on the first study, and hippocampal volume, tissue [NAA] and total hippocampal NAA were all significantly correlated with CAMCOG by the second study (Fig. 3). Exactly the same pattern was observed if the MMSE score was used instead of CAMCOG.

Fig. 3 Relationships between hippocampal [NAA] and total hippocampal NAA and cognitive scores. Study 2 was performed 1 year later than study 1. Filled squares, Alzheimer’s disease patients; filled triangles, control subjects. r = Spearman’s rho.

Hippocampal volume, tissue [NAA] and total hippocampal NAA were also significantly correlated with the NINCDS criteria on both studies, showing significant decreases as the NINCDS status progressed from negative through possible to probable (P < 0.01 in all cases). Of the other MRS‐derived parameters, only total hippocampal creatine showed a significant correlation with CAMCOG (P = 0.036), MMSE (P = 0.009) and NINCDS status (P = 0.027), but this presumably is a result of hippocampal atrophy, rather than a reduction in Cr concentration.

Left hippocampal volume correlated significantly with tissue [NAA] in the second study (P = 0.020). There was no significant correlation, however, for the first study or the right hippocampus, where both the volume changes and NAA reductions were smaller.

Discriminatory ability of MRS parameters

To determine the power of the MRS parameters to classify subjects correctly into cognitively impaired subjects (possible + probable Alzheimer’s disease) and controls, a linear discriminant analysis was performed. On its own, left hippocampal [NAA] gave a correct classification of 70% for the first study and 68% for the second study. Hippocampal volume gave slightly better discrimination (74 and 91%, respectively, for the two studies). The best classification was achieved by including left side [NAA] and left hippocampal volume, when 91% of cases were classified correctly (13 out of 14 controls and eight out of nine Alzheimer’s disease subjects). When the same classification was applied to the second study, 100% of cases were classified correctly. It was not possible to discriminate possible from probable Alzheimer’s disease or controls on the basis of the MRS parameters (40% of probables were classified as possible, and possible cases were assigned to all three groups). Limiting the analysis to control subjects and ‘probable’ Alzheimer’s disease, the combination of left side [NAA] and hippocampal volume gave correct classification of 95% of the cases in the first study (one control classified as ‘probable’) and 100% of cases in the second study. Right hippocampal [NAA] and volume gave poorer discrimination in all cases.


1H MR spectra were obtained and quantified successfully from hippocampi of elderly controls and patients with NINCDS probable and possible Alzheimer’s disease. The left hippocampus showed consistently larger differences between Alzheimer’s disease subjects and controls, and in addition showed larger changes over 1 year than did the right hippocampus. One can ask what additional information is afforded by MRS over MRI studies. It is as follows: a significantly lower concentration of NAA was found in the left hippocampus in Alzheimer’s disease subjects compared with controls. As NAA concentrations reported here have been corrected for tissue atrophy, this reduction did not simply reflect a smaller amount of tissue giving rise to the signal. This difference was maintained in a separate study on the same subjects 1 year later. If NAA is assumed to be exclusively in neurones, this result suggests that neuronal loss (or dysfunction) was disproportionately greater than general tissue atrophy in this region, demonstrating that the pathological process alters both tissue composition and overall hippocampal volume. The combination of these two parameters (hippocampal volume and [NAA]) gave better discrimination of Alzheimer’s disease subjects from controls than did either of the parameters separately, in agreement with the findings of Schuff et al. (1997).

The role of NAA is not yet entirely clear, and hence the interpretation of changes in tissue concentration is difficult. While it has been considered a neuronal marker, based on immunocytochemical evidence (Moffett et al., 1991; Simmons et al., 1991), it may also reflect mitochondrial function (Bates et al., 1996). The measured NAA concentration may therefore reflect both the total number of neurones in a given volume of tissue and the metabolic status of those neurones. We have measured not only the NAA concentration, but also the total amount of NAA in the hippocampus. The total amount of NAA may report the ‘functional status’ of the hippocampus, being affected by both the total number and the viability of the hippocampal neurones.

The finding of greater changes in the left hippocampus is interesting, as it parallels findings by other authors, although a complete consensus on hippocampal asymmetry in Alzheimer’s disease has not been reached. In many cases (e.g. Jack et al., 1992, 1997), only the average or total hippocampal volume is reported. In a study of elderly subjects with Alzheimer’s disease, mild dementia and normal cognition (Wolf et al., 2001), left hippocampal volume was found to discriminate the groups better than right hippocampal volume. Similarly, left hippocampal volume was the best predictor of recall of verbal information in Alzheimer’s disease subjects, while right hippocampal volume better predicted recall of spatial information (de Toledo‐Morrell et al., 2000). The left side of the brain generally is regarded as controlling verbal functions, while the right controls praxis. We suspect that one of the reasons that we and others find greater atrophy and metabolic changes in the left hippocampus in early Alzheimer’s disease could be a referral bias in the subjects, i.e. changes in verbal ability may be noticed sooner by patients and carers than a decline in practical ability, e.g. an increase in clumsiness, which may be seen as ‘normal’ in an elderly subject.

We did not see a significant decrease over 1 year in the hippocampal NAA concentration in the Alzheimer’s disease group, despite a significant decline in cognitive ability, as measured by the CAMCOG test (although the change in MMSE was not significant). There was a significant decrease in the left hippocampal volume in the Alzheimer’s disease subjects, though not in the controls, over this time. The tissue NAA concentration and the hippocampal volume were combined to give the total hippocampal NAA (in µmol). This parameter decreased in six out of the eight Alzheimer’s disease subjects (four out of five probable Alzheimer’s disease, and two out of three possible Alzheimer’s disease) although this was not statistically significant.

The hippocampus is a difficult part of the brain from which to obtain reliable spectra. It is close to areas of bone and air, both of which distort the magnetic field homogeneity, leading to broad linewidths, limiting spectral resolution and also making water suppression difficult. Other studies have reported failure of automated shimming at this site (Kantarci et al., 2000). The small volume of the hippocampus also meant that the signal‐to‐noise ratios of the metabolite spectra were low. This is exaggerated by substantial quantities of CSF in the voxel accompanying hippocampal atrophy. In this study, we found that the proportion of CSF within the 1HMRS voxel rose from typically 7% in the age‐matched controls to 15–25% in Alzheimer’s disease patients. Thus the volume of tissue within the 4.5 ml voxel was considerably reduced in some cases. In addition, we found that the signals from Cho, Cr and MI were often on a sloping baseline derived from incomplete water suppression, and so were difficult to quantify. The NAA signal, being further from the water frequency, was much easier to quantify reliably, which is reflected in the greater reproducibility of the NAA measurements from a single subject, as described in the Methods section.

In common with almost all previous reports, we have reported ratios rather than concentrations of MI. We did not calculate the absolute concentration of this metabolite since its signal arises from several non‐identical protons. The signal decays in a complicated manner during the pulse sequence because of magnetic interactions (J‐coupling) between the protons. Although the signal area is proportional to the concentration of MI for a given set of pulse sequence parameters, the proportionality constant is not straightforward to measure. This is in contrast to the signals from NAA, Cho and Cr, which arise from groups of identical protons and are not affected by J‐coupling. The signals from these metabolites decay exponentially during the pulse sequence, and the signal area can be simply related to the concentration.

The time course of the reduction in NAA concentration compared with the time course of Alzheimer‐pathology‐induced atrophy has yet to be clarified. If the NAA reduction precedes volume changes, then, in the future, with improved signal‐to‐noise ratio and availability of MRS on clinical MRI machines, there could be a role for NAA determination as an early diagnostic aid. This would probably require the comparison of spectra from two volumes of interest: the left hippocampus and a control region typically involved later in the disease, which is an approach similar to that adopted by Jessen et al. (2000). If, however, NAA reductions mirror the time course of grey matter atrophy, then it is likely that there will have to be significant improvements in signal‐to‐noise ratio for proton spectroscopy to be more than a research tool. If MRS does achieve clinical utility, rapid automatic, or semi‐automatic, calculation of NAA/Cr or NAA/(Cr + Cho) ratios would probably be required for widespread application, rather than the time‐consuming method of absolute quantification and atrophy correction presented in this study.

Will NAA measurement have a role as a surrogate marker in drug trials? The answer depends partly upon whether a potential drug will maintain NAA levels merely by slowing neuronal dysfunction or death, or whether the drug would also demonstrate some sort of additional NAA‐elevating action. If it is only the former, then it is likely that signal‐to‐noise ratios will need to be improved for hippocampal MRS to have any significant role in drug trials. Our experience suggests that MI may be less valuable in Alzheimer’s disease due to the greater difficulty involved in its measurement.

In summary, we have detected reductions in [NAA] in the hippocampus, a region of the brain which promises to be the most sensitive indicator of early pathology in Alzheimer’s disease. The methodology that we have employed corrects for atrophy within the spectroscopic voxel. Furthermore, by employing concurrent volumetric assessment of the hippocampus, total hippocampal NAA could be calculated. With the emergence of novel therapies designed to reduce the progression of the pathological processes in Alzheimer’s disease, it will become increasingly important to detect the presence of the disease at the earliest possible time point.

Additional studies exploiting hippocampal MRS are required to determine the viability of MRS as a surrogate marker in this disease.


We wish to thank Dr Andrew Blamire and Dr Bheeshma Rajagopalan for helpful advice and discussions. The Medical Research Council, UK (Foresight Challenge grant 207) and Bristol‐Myers Squibb provided financial support for this study.


View Abstract