Low Striatal Glutamate Levels Underlie Cognitive Decline in the Elderly: Evidence from In Vivo Molecular Spectroscopy

Abstract

Glutamate (Glu), the principal excitatory neurotransmitter of prefrontal cortical efferents, potentially mediates higher order cognitive processes, and its altered availability may underlie mechanisms of age-related decline in frontally based functions. Although animal studies support a role for Glu in age-related cognitive deterioration, human studies, which require magnetic resonance spectroscopy for in vivo measurement of this neurotransmitter, have been impeded because of the similarity of Glu's spectroscopic signature to those of neighboring spectral brain metabolites. Here, we used a spectroscopic protocol, optimized for Glu detection, to examine the effect of age in 3 brain regions targeted by cortical efferents—the striatum, cerebellum, and pons—and to test whether performance on frontally based cognitive tests would be predicted by regional Glu levels. Healthy elderly men and women had lower Glu in the striatum but not pons or cerebellum than young adults. In the combined age groups, levels of striatal Glu (but no other proton metabolite also measured) correlated selectively with performance on cognitive tests showing age-related decline. The selective relations between performance and striatal Glu provide initial and novel, human in vivo support for age-related modification of Glu levels as contributing to cognitive decline in normal aging.

Introduction

Glutamate (Glu) is a key molecule in cellular metabolism, the most abundant excitatory neurotransmitter in the mammalian nervous system, and the principal neurotransmitter of cortical efferents (Fonnum et al. 1981; Thangnipon et al. 1983). Altered Glu metabolism or neurotransmission may contribute to the neural mechanisms underlying age-related declines in behavioral functions served by cortical efferent systems. Independent studies describe alterations to the glutamatergic system with age and suggest a role for Glu in cognitive functioning, but few animal and no human studies have demonstrated the functional significance of age-related alterations to the glutamatergic system. Autoradiography of postmortem rodent and human tissue demonstrates lower Glu receptor binding with older age in basal ganglia (Mitchell and Anderson 1998; Villares and Stavale 2001), cerebellum (Tsiotos et al. 1989; Simonyi et al. 2000), and hippocampus (Tamaru et al. 1991; Court et al. 1993). Chromatrographic assays detect age-related declines in total Glu concentration in homogenized rat cerebrum (Tyce and Wong 1980; Dawson et al. 1989; Benuck et al. 1995) but not autopsied putamen (Kornhuber et al. 1993) or biopsied basal frontal, basal temporo-occipital, temporal, and parietal cortices (Knorle et al. 1997) of human tissue.

Behavioral pharmacology has provided the mainstay of evidence for Glu's role in cognition (Robbins and Murphy 2006). In rodents, blockade of Glu receptors impairs spatial working memory (Morris et al. 1986) and object recognition (Winters and Bussey 2005). In healthy human volunteers, glutamatergic antagonists (e.g., ketamine) impair performance on tests of verbal (Parwani et al. 2005) and nonverbal declarative memory (Newcomer et al. 1999), verbal fluency, and problem solving (Krystal et al. 1999).

Evidence for age-linked cognitive changes to the glutamatergic system derives primarily from correlations between rodent behavior and Glu receptor levels measured with in vitro autoradiography. For example, age-related decreases in N-methyl-D-asparate (NMDA) receptor binding in the striatum correlate with better spatial learning (Nicolle et al. 1996) and attentional set shifting (Nicolle and Baxter 2003).

In vivo human studies require estimates of Glu from magnetic resonance spectroscopy (MRS) examination. Heretofore, MRS of Glu has been difficult to conduct because Glu's chemical structure gives rise to multiple resonances that overlap with signals from other brain metabolites. Noninvasive, in vivo measurement of Glu would enable testing of specific hypothesis about the regional distribution of Glu, its relation to aging, and its contribution to cognitive, motor, and sensory functioning. In light of new evidence that modulation of Glu receptors may ameliorate symptoms of schizophrenia (Patil et al. 2007), the ability to measure Glu in living tissue is especially relevant, will allow for the longitudinal tracking of drug efficacy and contribute to understanding the pathogenesis of other neurological diseases involving altered glutamatergic neurotransmission, including multiple sclerosis (Srinivasan et al. 2005), Parkinson's disease (Kuiper et al. 2000), Alzheimer's disease (Greenamyre 1991), and bipolar disorder (Beneyto et al. 2007).

Here, we used an in vivo MRS method optimized to detect Glu (Mayer and Spielman 2005) in young and elderly, healthy men and women. Our aim was to dissociate the biochemical profile of 3 brain regions targeted by prefrontal glutamatergic efferents and showing differential effects of age on gross morphology: striatum (Jernigan et al. 1991; Raz et al. 2003), cerebellum (Sullivan et al. 2000; Jernigan et al. 2001), and pons (Raz et al. 2001; Sullivan et al. 2004). Simultaneous measurement of N-acetylaspartate (NAA), total creatine (tCr), choline-containing compounds (Cho), and myo-inositol (mI) permitted testing of the selectivity of Glu over other resonances in predicting performance on cognitive tests sensitive primarily to frontostriatal systems integrity and secondarily to frontopontocerebellar system recruitment when primary systems are compromised (Allen and Courchesne 2003; Desmond et al. 2003; Sullivan 2003).

Materials and Methods

Participants

Volunteers were 12 young (range = 19–33 years) and 12 elderly (range = 67–84 years), right-handed, nonsmoking healthy men and women, recruited from the local community (Table 1). All subjects provided signed, informed consent to participate in this study, which was approved by the Institutional Review Boards of SRI International and Stanford University. All participants underwent a thorough psychiatric interview by a trained research psychologist using the Structured Clinical Interview for the Diagnostic and Statistical Manual (DSM) IV to detect DSM-IV diagnoses or medical conditions that can affect brain functioning (e.g., diabetes, head injury, epilepsy, uncontrolled hypertension, radiation/chemotherapy) or preclude MR study (e.g., pacemakers). The 2 groups did not differ significantly in education or estimated general intelligence, although the elderly had a higher socioeconomic status than the young (Hollingshead 1975). The elderly scored lower than the young on the Dementia Rating Scale (DRS) (cutoff for dementia ≤124 out of 144; Mattis 1988) but well within the normal range of previously published values for healthy elderly individuals living in the community (e.g., 137.48 ± 5.18; Vitaliano et al. 1984). The elderly had significantly greater body mass index (BMI) than the young (slightly overweight, i.e., BMI between 25 and 29.9 kg/m²) but close to the mean BMI calculated from the 5200 subjects participating in the cardiovascular health study (26.3 ± 3.9 kg/m²; Janssen et al. 2005). The elderly had significantly higher systolic blood pressure than the young group but all were in the prehypertensive range (i.e., between 120 and 139 mmHg; Chobanian et al. 2003).

Table 1.

Group demographics: means, standard deviations, and ranges

	Age (years)	Education (years)	NARTb IQ	SESc (self)	DRS	BMI	Systolic BP (mm Hg)	Diastolic BP (mm Hg)
Young (6 M, 6 F)	25.50 ± 4.34 (19–33)	16.25 ± 2.18 (14–21)	113.67 ± 5.66 (102–121)	26.08 ± 10.11 (11–47)	141.42 ± 2.57 (136–144)	23.08 ± 12.98 (19–29)	114.00 ± 16.16 (93–143)	68.58 ± 8.46 (59–88)
Elderly (6 M, 6 F)	77.67 ± 4.94 (67–84)	17.08 ± 1.98 (14–21)	118.17 ± 5.59 (107–126)	18.33 ± 6.24 (11–29)	137.25 ± 4.71 (130–143)	26.64 ± 4.55 (21–36.8)	135.9 ± 221.84 (108–178)	75.42 ± 11.65 (60–96)
P valuea	<0.0001	0.337	0.063	0.034	0.013	0.034	0.011	0.114

^at-tests. ^bNational Adult Reading Test, ^cSocio-economic status.

MRS Acquisition

The MRS was performed using constant time point–resolved spectroscopy (CT-PRESS; Dreher and Leibfritz 1999) on a gradient-echo 3-T MR scanner. Single voxels were manually positioned in left or right striatum (10.6 cc), left or right cerebellum (9.8 cc), and central pons (5.9 cc; Fig. 1); hemisphere of voxel placement was balanced across subjects and groups. The acquisition time was ∼9 min per voxel (average echo time [TE] = 139, chemical shift [CS] encoding steps, Δt₁/2 = 0.8 ms, repetition time [TR] = 2 s, 2 averages; Mayer and Spielman 2005). A scan without water suppression was acquired (17 CS encoding steps, Δt₁/2 = 6.4 ms, 2 averages) to measure the tissue water content used to normalize the metabolite signal intensities. Data acquired without water suppression aided in metabolite quantification and were apodized in t₂ with a 5-Hz Gaussian line broadening and zero filled up to 4K points for each TE.

Figure 1.

Voxel location, sample, and average spectra for 3 brain regions. (A) Voxel location in each region. (B) Sample spectra from a young (23 years, blue) and elderly (75 years, red) man. (C) Average signal intensity of metabolites relative to tissue water (striatum, n [young] = 11, n [old] = 11; cerebellum, n [young] = 11, n [old] = 8; pons, n [young] = 11, n [old] = 11), *P ≤ 0.05.

After performing a fast Fourier transform (FFT; Neil et al. 1998) along t₂, water spectra were evaluated by peak integration. The amount of cerebral spinal fluid (CSF) and tissue water was estimated by fitting the data across the 17 TEs to a biexponential model (Mayer et al. 2007). Apodization of the water-suppressed data entailed multiplication with sine-bell functions in both time dimensions and zero filling up to 4K × 1K data points. After performing a 2-dimensional (2D) FFT, effectively decoupled 1D CT-PRESS spectra were obtained by integrating the 2D spectrum in magnitude mode along f₂ within a ±13-Hz interval around the spectral diagonal. Metabolite signals in the 1D spectra were determined by peak integration with an interval of ±6 Hz. The quality of the spectra allowed evaluation of signals of the major proton metabolites: NAA (2.01 ppm), tCr (3.03 and 3.93 ppm), Cho (3.24 ppm), mI (3.58 ppm), Glu (2.36 ppm), and glutamate + glutamine (3.78 ppm; Fig. 1, Table 2). Data from 2 striatal (1 young, 1 elderly), 2 pontine (2 elderly), and 5 cerebellar voxels (1 young, 4 elderly), representing 7 subjects were excluded because of poor spectral quality.

Table 2.

Metabolite levels in arbitrary units: means and standard deviations

			NAA	Glu	tCr	Cho	mI	Glutamate + glutamine
Striatum	Young	n = 11	0.123 ± 0.015	0.019 ± 0.003	0.074 ± 0.005	0.061 ± 0.004	0.009 ± 0.003	0.014 ± 0.003
	Elderly	n = 11	0.110 ± 0.017	0.013 ± 0.003	0.066 ± 0.009	0.060 ± 0.011	0.011 ± 0.003	0.013 ± 0.001
		P valuea	0.069	0.0002	0.015	0.829	0.306	0.152
Cerebellum	Young	n = 11	0.158 ± 0.013	0.012 ± 0.005	0.134 ± 0.009	0.116 ± 0.012	0.024 ± 0.005	0.016 ± 0.004
	Elderly	n = 8	0.164 ± 0.015	0.016 ± 0.006	0.152 ± 0.019	0.111 ± 0.017	0.019 ± 0.004	0.018 ± 0.004
		P value	0.338	0.122	0.013	0.499	0.038	0.513
Pons	Young	n = 12	0.230 ± 0.024	0.016 ± 0.003	0.083 ± 0.008	0.146 ± 0.013	0.016 ± 0.005	0.016 ± 0.007
	Elderly	n = 10	0.237 ± 0.014	0.017 ± 0.006	0.091 ± 0.007	0.166 ± 0.020	0.017 ± 0.005	0.018 ± 0.004
		P value	0.434	0.891	0.016	0.016	0.749	0.611

^at-tests.

Magnetic Resonance Imaging Acquisition

An axial fast spin echo magnetic resonance imaging (field of view = 24 cm, frequency encode = 256, TE1/TE2/TR = 17/102/7500 ms, phase encode = 192, echo train length = 8, slice thickness = 2.5 mm, 0 spacing) was used to quantify tissue and CSF volumes in each voxel. Signal-to-noise ratio was adequately robust to determine fractions of gray matter and white matter with an automatic segmentation routine in the striatal voxel (Lim and Pfefferbaum 1989).

Neuropsychological Assessment

Cognitive testing focused on component processes associated with prefrontal–striatal systems and included measures of fluency, working memory, set shifting, and rule formation. “Phonological fluency” required subjects to say, in 1 min per letter, words beginning with the letter F, A, and then S (Borkowski et al. 1967). “Semantic fluency” required subjects to say words, in 1 min per category, that were inanimate objects, animals, and alternating birds and colors (Newcombe 1969). The “figural fluency test” required that subjects draw unique designs following rules and ignoring distractors (Ruff et al. 1987). Tests of working memory were “forward and backward digit and block spans” and the modified “Sternberg variable memory load test,” for which subjects remembered 1, 3, or 6 letters over 5 s intervals; score was reaction time (RT) with the highest load, minus RT with lowest load when the stimulus was present (Desmond et al. 1997, 2003).

Set shifting was assessed with the “intra–extra dimensional shift task” of the Cambridge Neuropsychological Test Automated Battery (CANTAB; http://www.camcog.com), requiring subjects to discriminate between 2-alternative, forced-choice stimuli. Subjects first focused attention on specific attributes of compound stimuli (i.e., intradimensional shift) and then shifted attention to a previously irrelevant stimulus dimension (i.e., extradimensional shift). The score was the number of trials to reach criterion in the extradimensional stage minus the number of trials to reach criterion in the intradimensional stage.

The CANTAB “Stockings of Cambridge” component evaluated spatial planning abilities. Subjects were shown 2 displays, each containing 3 colored balls within stockings and were asked to move the balls by tapping the touch-sensitive screen in the lower display to copy the goal pattern shown in the upper display, using as few moves as possible. The score was the number of problems solved in the minimum number of moves.

Results

Glu in the Striatum Is Lower in the Elderly than the Young Group

A 2-group (young vs. elderly), repeated measures (3 regions and 6 metabolites) analysis of variance revealed 2 significant interactions: group-by-region (F_1,2 = 5.500, P_GG = 0.0097) and group-by-region-by-metabolite (F_1,10 = 4.986, P_GG = 0.0013). The 3-way interaction indicated a significant age effect. Follow-up comparisons indicated that Glu in the striatum was lower in the elderly than the young group (P = 0.0002). tCr levels were lower in the striatum (P = 0.015) but higher in the cerebellum (P = 0.013) and pons (P = 0.016) of the elderly than the young. Cho was higher in the pons (P = 0.016), and mI was lower in the cerebellum (P = 0.038) of the elderly than the young (Fig. 1, Table 2).

Volumes of the striatal voxel segmented by tissue type indicated that the elderly had a smaller volume of gray matter (P = 0.0193) and larger volumes of CSF (P = 0.0237) and white matter (P = 0.0366) than the young. Higher levels of striatal NAA (r = 0.46), Glu (r = 0.60), and tCr (r = 0.54) correlated with larger gray matter volumes in the 2 age groups combined.

Elderly Perform Worse than the Young on Selective Frontal/Executive Tasks

Performance on cognitive tests was significantly poorer in the elderly than younger group on semantic and figural fluency, working memory, set shifting, and spatial planning (Table 3). The elderly produced a similar number of words beginning with F, A, or S in phonological fluency but listed significantly fewer animals and inanimate objects (F_1,23 = 8.13, P = 0.009) and bird/color alterations (F_1,23 = 11.48, P = 0.003) and drew fewer total unique designs (F_1,23 = 25.47, P = 0.0004) than the young. Age minimally affected forward and backward digit span but significantly affected forward (F_1,23 = 17.46, P = 0.0004) and backward (F_1,23 = 8.81, P = 0.007) block span. In the Sternberg memory task, the elderly had significantly slower RTs than the young (F_1,23 = 10.80, P = 0.003). Whereas the elderly group was successful in intradimensional attention shifting, they performed significantly worse than the young in extradimensional attention shifting in the intra–extra dimensional shift task (F_1,23 = 14.30, P = 0.001). On the Stockings of Cambridge task, the older group solved fewer problems in the minimum number of moves than the younger group (F_1,23 = 8.80, P = 0.007).

Table 3.

Mean ± standard deviation of raw scores on neuropsychological measures

	Young (n = 12)	Elderly (n = 12)	P valuea
Fluency
Phonological fluency	40.33 ± 11.74	39.75 ± 12.90	0.950
Semantic fluency (A + O)	48.92 ± 7.63	38.58 ± 10.10	0.009
Semantic fluency (B/C)	17.92 ± 3.42	13.00 ± 3.98	0.003
Figural fluency	100.25 ± 13.44	80.33 ± 9.76	<0.001
Working memory
Digit forward	8.75 ± 2.05	8.92 ± 2.11	0.822
Digit backward	7.25 ± 2.49	7.00 ± 2.37	0.918
Blocks forward	9.33 ± 1.23	7.33 ± 1.07	<0.001
Blocks backward	9.00 ± 1.28	7.33 ± 1.67	0.007
Sternbengb	259.10 ± 155.67	435.06 ± 97.36	0.003
Set shifting
Intra–extra dimensional shift taskb	5.08 ± 3.59	22.63 ± 3.59	0.001
Spatial planning
Stockings of Cambridge	10.25 ± 1.42	8.17 ± 221	0.007

^at-tests.

^bHigher scores = worse performance.

Glu in the Striatum Predicts Cognitive Performance

Simple regression analysis included all subjects to test the relations between cognitive performance and each of the 4 metabolites showing age effects and revealed a number of significant correlations. To adjust alpha for multiple comparisons, we calculated a family-wise Bonferroni P value for 5 comparisons, representing the 5 metabolites measured. Because the direction of the correlations was predicted, 1-tailed P values were used. The resulting correction required for α = 0.05 was P = 0.02. Higher levels of striatal Glu correlated with better performance on semantic and design fluency, forward and backward block span, and the Sternberg and the intra–extra dimensional shift tasks (Fig. 2).

Figure 2.

Scatter plots of simple regressions between striatal Glu and neuropsychological measures. Positive correlations with Glu levels were expected for neuropsychological measures where higher scores indicate better performance (fluency and working memory tests), and negative correlations with Glu levels were expected for neuropsychological measures (Sternberg and CANTAB tests) where lower scores indicate better performance.

Selectivity of these simple brain metabolite/performance correlations was tested with multiple regression using 3 approaches. The first set of analyses examined the selectivity of striatal Glu against Glu in pons and cerebellum in predicting performance levels. Glu in the 3 brain regions combined accounted for 48–58% of the variance in predicting performance on semantic and design fluency and nonverbal working memory (forward and backward block span, Table 4). The independent contribution from striatal Glu was greater than that from the pons, which was greater than that from the cerebellum; neither pontine nor cerebellar Glu were significant unique predictors of performance. In addition, Glu in all 3 regions together accounted for 53% of the variance of intra–extra dimensional shift task performance. In contrast to the observed selective contributions from striatal Glu to fluency and working memory, Glu in each of the 3 brain regions made a significant independent contribution to intra–extra dimensional shift task performance.

Table 4.

Correlations between Glu in 3 regions and neuropsychological measures

	All 3 regions	r2	Striatum	Cerebellum	Pons
	P value		P value	P value	P value
Semantic fluency (A + O)	0.015	0.543	0.003	0.480	0.685
Semantic fluency (B/C)	0.032	0.481	0.004	0.496	0.031
Figural fluency	0.029	0.488	0.007	0.121	0.088
Blocks forward	0.009	0.575	0.002	0.681	0.038
Blocks backward	0.030	0.486	0.007	0.135	0.080
Intra–extra dimensional shift task	0.018	0.526	0.020	0.011	0.042

The second analysis approach addressed whether age-related decline in striatal gray matter volume was responsible for the relationships between Glu and performance. Accordingly, we conducted multiple regression analyses, using striatal gray matter volume and Glu as simultaneous predictors of performance. These analyses identified striatal Glu over gray matter as the unique predictor of performance on semantic (Glu P = 0.0312, gray matter P = 0.8506) and design fluency (Glu P = 0.0484, gray matter P = 0.2926) and on forward (Glu P = 0.0135, gray matter P = 0.2654) and backward (Glu P = 0.0203, gray matter P = 0.4366) block spans. Although NAA did not show an age effect, we examined its relationship with test performance because NAA is considered a marker of neuronal integrity and thus a likely predictor of performance. In contrast to the analysis based on Glu and gray matter volume, multiple regression analyses using striatal gray matter volume and NAA as performance predictors revealed gray matter (P values ranged from 0.0074 to 0.0414) rather than NAA levels as the unique contributor to performance on the tests of fluency and working memory.

The third set of analyses focused on the striatum because of the many significant correlations between striatal Glu and performance and examined the selectivity of Glu among the other major metabolites (NAA, tCr, Cho, mI) in predicting performance. For the tests showing simple correlations with striatal Glu (i.e., semantic and design fluency, forward and backward block span, and intra–extra dimensional shift task), Glu endured as a significant independent predictor of performance over and above the contribution from the remaining 4 metabolites (Table 5).

Table 5.

Correlations between striatal metabolites and neuropsychological measures

	All 5 metabolites	r2	Glu	NAA	tCr	Cho	mI
	P value		P value	P value	P value	P value	P value
Semantic fluency (A + O)	0.146	0.376	0.062	0.643	0.775	0.834	0.376
Semantic fluency (B/C)	0.011	0.578	0.002	0.821	0.500	0.679	0.025
Figural fluency	0.022	0.531	0.005	0.939	0.838	0.116	0.124
Blocks forward	0.018	0.543	0.004	0.548	0.445	0.349	0.567
Blocks backward	0.018	0.544	0.001	0.536	0.104	0.215	0.573
Intra–extra dimensional shift task	0.080	0.433	0.020	0.184	0.594	0.841	0.072

Discussion

These results provide novel, in vivo evidence for human age-related decline in striatal Glu that contributes selectively to cognitive functions supported by frontostriatal systems. Prior to this study, the relevance of decreasing regional Glu as a mechanism of age-related functional decline has been underappreciated because of the difficulty in measuring Glu in vivo in humans.

Glu levels referenced to tissue water content showed differential regional age effects, with lower Glu in the striatum, but not pons or cerebellum, in elderly than younger healthy individuals. This observation is consistent with other human MRS studies, reporting lower Glu concentrations in older age in the hippocampus, anterior cingulate cortex (Schubert et al. 2004), motor cortex (Kaiser et al. 2005), frontal white matter, parietal gray matter, and basal ganglia (Sailasuta et al. 2006). Rodent studies also note an age-related decline in total Glu concentration measured in whole-brain extracts using either scintillation counting of injected radioactive glucose (de Koning-Verest 1980; Tyce and Wong 1980) or chromatography (Dawson et al. 1989; Benuck et al. 1995).

The other 4 major proton metabolites measured showed varying patterns of aging. When expressed relative to tissue water, tCr was lower in the striatum but higher in the cerebellum and pons of the elderly than the young group. Increases in tCr with age have been reported for whole-brain gray matter (Pfefferbaum et al. 1999), frontal gray matter (Chang et al. 1996), parietal white matter (Saunders et al. 1999; Leary et al. 2000; Schuff et al. 2001), and the pons (Costa et al. 2002). However, absences of change to tCr in frontal and parietal gray matter (Lundbom et al. 1999; Schuff et al. 2001) and decreases to tCr in the striatum (Charles et al. 1994) with age have also been reported.

The use of water as a referent in the present study permitted examination of age effects on tCr, in contrast to traditional studies in which tCr is used as a referent. tCr is frequently used as a referent because, with some exceptions (Chang et al. 1996; Pfefferbaum et al. 1999; Meyerhoff et al. 2004), it is considered to be robust to change with age or disease. In the present study, Glu was observed to be lower in the older than younger adults whether referenced to water or tCr. That is, despite lower levels of tCr in the striatum of the elderly, the ratio of Glu/tCr in the elderly (0.198 ± 0.037) was significantly lower than in the young (0.253 ± 0.033, P = 0.0015) group. Likewise, the groups did not differ significantly whether Glu was referenced to water or tCr (cerebellum: elderly 0.109 ± 0.035, young 0.091 ± 0.039, P = 0.3134; pons: elderly 0.182 ± 0.067, young 0.199 ± 0.039, P = 0.4842), despite the higher tCr levels in these regions of the elderly.

NAA varied regionally but showed no age effect when referenced to tissue water. Several studies report little or no age-related decline in NAA of healthy men and women and have examined various brain regions (Kwo-On-Yuen et al. 1994; Pfefferbaum et al. 1999; Saunders et al. 1999; Adalsteinsson et al. 2000; Grachev and Apkarian 2000; Sijens et al. 2003), including frontal white (Kaiser et al. 2005) and gray matter (Chang et al. 1996), parietal white matter (Meyerhoff et al. 1994; Leary et al. 2000), and pons (Costa et al. 2002). This lack of age effect occurred regardless of whether metabolites levels were expressed in absolute values (concentrations or institutional units) or as ratios relative to other metabolites. Absence of NAA difference between age groups (Pfefferbaum et al. 1999) or decline over time (Adalsteinsson et al. 2000) suggests that the number of viable neurons per unit of gray matter is preserved with age.

Cho in the pons was greater in the elderly than the young subjects, a finding consistent with an earlier study (Moreno-Torres et al. 2005). Higher Cho in the elderly group (Soher et al. 1996; Pfefferbaum et al. 1999; Angelie et al. 2001) occurred in frontal and parietal white matter (Moats et al. 1994; Chang et al. 1996; Leary et al. 2000; Kaiser et al. 2005) and frontal and occipital gray matter (Moats et al. 1994; Chang et al. 1996). Because an elevated Cho signal has been observed in multiple sclerosis (Srinivasan et al. 2005) and is a marker for tumor activity (Star-Lack et al. 2000), both associated with increased cell membrane turnover or breakdown, these results may reflect increased Cho release related to membrane breakdown in the aging process.

The only age-related difference observed for mI was lower levels in the cerebellum of the elderly. This finding is difficult to interpret because mI is typically higher in older age (Chang et al. 1996; Ross et al. 2006); but see (Kaiser et al. 2005) suggesting increased gliosis in aging (Brand et al. 1993). Given that 4 of the 12 elderly subjects had unusable cerebellar spectral data, the usable subject sample may have been biased toward the individuals who were aging most successfully.

In the combined age groups, Glu in the striatum correlated selectively with performance on executive tests also showing age-related decline: semantic and figural fluency, working memory, and set shifting. Support for these tests as tapping frontal executive systems originally derives from studies of patients with frontal lobe excisions (Goldman-Rakic 1996; Robbins 1996) and of patients with Parkinson's disease, whose primary pathology is striatal, and who present with diminished verbal (Milner 1963; Jacobs et al. 1995; Troster et al. 1998) and nonverbal (Jones-Gotman and Milner 1977; Lacritz et al. 2000) fluency, poor working memory (Sullivan and Sagar 1989, 1991; Sullivan et al. 1993; Lewis et al. 2003), and compromised set shifting (Milner 1963; Portin et al. 1984; Owen et al. 1993; Hochstadt et al. 2006; Monchi et al. 2007). Indeed, executive functions require the integrity of neuronal circuits linking multiple structures, rather than the integrity of isolated structures along that circuit (Sullivan 2003; Owen 2004; Knight 2007). The current results support a role for the striatum, as a target of prefrontal cortical glutamatergic fibers, in mediating cognitive function. These results concur with primate studies demonstrating that lesions of the portion of the caudate to which the lateral orbitofrontal cortex projects result in perseverative errors in behavioral set shifting (Divac et al. 1967).

An intrinsic age-related loss of Glu in the striatum, likely resulting in reduced Glu receptor activation and correlating with impaired cognitive performance, is consistent with literature describing cognitive impairment as a result of NMDA Glu receptor inhibition in normal healthy individuals (Krystal et al. 1999; Newcomer et al. 1999; Parwani et al. 2005). These results do not contradict findings in other age-related diseases, such as Alzheimer's disease, where pathology may include increased Glu levels (Hoyer and Nitsch 1989; Noda et al. 1999) or increased NMDA receptor sensitivity to Glu (Wu et al. 1995) that may be corrected by antagonists, such as memantine. Furthermore, in contrast to the NMDA receptor channel blockers that can impair cognitive function (e.g., ketamine), memantine has faster kinetics (Rogawski et al. 1991; Lipton 2005), which may explain its therapeutic benefit and tolerability.

Among the metabolites measured, the most consistent and selective predictor of cognitive performance in the present study was striatal Glu. Notably, Glu superceded NAA, which is typically considered a marker for neuronal integrity (Griffin et al. 2002; Kaiser et al. 2005), as a significant correlate of performance. Commonly, instances of NAA as a correlate of performance have derived from studies of disease rather than health. For example, patients with Alzheimer's disease show significant correlations between mini mental state examination scores and NAA/Cr ratios (Jessen et al. 2001; Waldman and Rai 2003; Frederick et al. 2004) and DRS performance and NAA gray matter concentrations (Adalsteinsson et al. 2000). Because Glu has been elusive to quantitation, it may have been overlooked as a relevant marker of function, perhaps even more so than NAA, which is the most prominent signal in the proton spectrum of the healthy human brain and measured most frequently. Although one could argue that lack of age-related decline in NAA attenuates its contribution to performance, striatal gray matter volume, like Glu, did show an age effect yet was not a significant correlate of function.

An important consideration in the interpretation of these results is that the MRS-detectable Glu signal does not discriminate between the metabolic and neurotransmitter pool of Glu. It has been estimated that 70–80% of tissue Glu is present in the metabolic pool and 20–30% in glutamatergic nerve terminals (Fonnum 1993). The MRS-detectable decreases in Glu content may be a consequence of a change in metabolic activity reflecting decreased function or viability of neurons because Glu, like NAA, is located primarily in neurons (Ottersen et al. 1992). An argument against this interpretation is the absence of a concurrent age-related decline in NAA. Moreover, recent ¹³C-MRS studies suggest that majority of energy related to Glu metabolism is used to support events associated with Glu neurotransmission (Sibson et al. 1998; Rothman et al. 2003). Additionally, although the sample size is relatively small, the effect size, estimated with Cohen's d, which in this case = 2 for Glu in the striatum, was sufficiently large to justify the use of 24 subjects.

Selectivity of Glu in striatum as a meaningful predictor of the measured cognitive functions was supported by demonstrating its unique contribution to performance relative to Glu in the pons and cerebellum. The one exception to the brain regional selectivity of Glu was for the intra–extra dimensional shift task, for which better performance correlated with higher Glu levels in all 3 brain regions. This pattern of relations may be indicative of task complexity and the need to invoke the frontopontocerebellar system, in addition to the frontostriatal system, to accomplish the task (Desmond et al. 2003; Sullivan 2003; Raz and Rodrigue 2006; Huang et al. 2007).

This study presents initial in vivo quantification of regional levels of Glu in the human brain and its functional relevance. The observed selective relations between striatal Glu and performance on tests of verbal and nonverbal fluency, working memory, and set shifting, known to be sensitive to frontostriatal systems integrity, provide in vivo evidence for waning Glu in the striatum as a mechanism of age-related decline in selective functions supported by frontostriatal systems. These metabolite–function relations contribute to the validation of MRS-derived Glu as an in vivo marker of this neurotransmitter and support the use of in vivo molecular spectroscopy in investigating the involvement of Glu in mechanisms of cognitive decline in normal aging and in the pathophysiology of neuropsychological and neuropsychiatric disorders.

Funding

National Institute on Aging (AG017919 to E.V.S); National Institute on Alcohol Abuse and Alcoholism (AA012388 and AA005965 to A.P.).