Abstract

Methods

Results

Table 1 lists the demographic data of the subjects as well as the load of cerebrovascular disease in each group, evaluated by WMH and number of lacunes. Patients with SIVD were not markedly younger than the other groups (F = 2.2, df = 4,103, p < 0.08). MMSE scores of patients were comparable between AD and SIVD (F = 0.02, df = 2,53, p > 0.9), but substantially lower than the scores of CN subjects (F = 147, df = 1,106, p < 0.001), as expected. CN subjects with and without lacunes did not differ with respect to MMSE scores (F = 0.2, df = 1,50, p > 0.6). Sex was reasonably balanced in the AD and CN groups (all p > 0.5 by χ² tests), but women were under-represented in the SIVD group (p = 0.03).

In analyzing MRSI, we first determined whether metabolite differences between AD+L and AD−L patients and between CN+L and CN−L subjects needed to be considered. MANOVA tests indicated that there were no differences in MRSI data between AD+L and AD−L (p > 0.5) and between CN+L and CN−L (p > 0.9). Therefore, comparisons between SIVD and the other groups were performed without differentiating between individuals with and without lacunes. Table 2 lists [NAA] concentrations (in arbitrary units) in medial temporal lobe, including the hippocampus, in SIVD, AD, and CN. [NAA] losses in medial temporal lobe were strongly associated with diagnosis (p < 0.001). The effect remained significant after accounting for potential bias due to undersampling of small hippocampi, stemming from partial volume limitations of MRSI. However, 11% lower [NAA] in medial temporal lobe of patients with SIVD compared to CN subjects was not significant in post hoc tests to explain the main effect of diagnosis. In contrast, 21% lower [NAA] in patients with AD compared to CN subjects explained the main effect of diagnosis on medial temporal lobe [NAA] (left: F = 19.9, df = 2,82, p < 0.001; right: F = 29.6, df = 2,82, p < 0.001). Despite substantial [NAA] loss in AD but moderate [NAA] loss in SIVD relative to CN, differences between patients with AD and patients with SIVD in medial temporal lobe were not significant. For the numbers of patients tested and 80% power, the minimum detectable difference for [NAA] in medial temporal lobe was 16%.

Outside the medial temporal lobe in frontal and parietal lobe regions, cortical [NAA] was also associated with diagnosis (p < 0.003). [NAA] concentrations of frontal and parietal lobes in the different groups are listed in table 2. Compared to CN, patients with SIVD had about 18% lower cortical [NAA] in the frontal lobe (left: F = 10.7, df = 2,62, p < 0.002; right: F = 10.3, df = 2,62, p < 0.002) and about 27% lower cortical [NAA] in the parietal lobe (left: F = 36.2, df = 2,62, p < 0.0001; right: F = 12.6, df = 2,62, p < 0.001). This contrasts with results from the medial temporal lobe, where no significant [NAA] differences between SIVD and CN subjects were found. Compared to AD, patients with SIVD had about 13% lower cortical [NAA] in both left and right frontal lobes (F = 6.0, df = 2,50, p < 0.02) and, in addition, 20% lower cortical [NAA] in the left parietal lobe (F = 10.9, df = 2,50, p < 0.002), whereas differences in the right parietal cortex were not significant. In contrast to SIVD, patients with AD had no cortical [NAA] loss in the frontal lobe (both left and right p > 0.1) when compared to CN subjects. In the parietal lobe, however, patients with AD had—like SIVD—about 13% lower cortical [NAA] than CN subjects (left: F = 7.2, df = 2,91, p < 0.01; right: F = 5.6, df = 2,91, p < 0.01). [NAA] of white matter was not different between groups (all p > 0.2). For the numbers of subjects tested and 80% power, the minimum detectable difference of [NAA] in white matter was 13%.

Neither [Cho] nor [Cr] of medial temporal lobe was related to diagnosis (p > 0.4). Outside the medial temporal lobe, there was a trend for lower cortical [Cho] (p = 0.03) as a function of diagnosis. This was driven by lower cortical [Cho] in the left parietal lobe in SIVD relative to AD (F = 10.6, df = 2,42, p < 0.002) and to CN (F = 13.1, df = 2,53, p < 0.001) (data not shown). Variations of cortical [Cr] were not related with diagnosis.

We also tested [NAA], [Cho], and [Cr] of WMH and normal-appearing white matter for differences in 11 patients with SIVD, 17 patients with AD, and 11 CN subjects, who had WMH that were large enough to place MRSI voxels (approximately 1 cm³ volume). There was no significant effect of diagnosis on [NAA] variations in WMH. Across groups, however, [NAA] in WMH was lower by about 15% (F = 11.3, df = 2,37, p < 0.005) than in normal-appearing white matter. There was also no significant effect of diagnosis on [Cho] and [Cr] variations in WMH and neither [Cho] nor [Cr] differed significantly between WMH and normal-appearing white matter.

To test whether AD and SIVD are associated with different regional patterns of [NAA] loss, we transformed [NAA] values into z-scores to account for systematic technical differences between PRESS and multislice MRSI, such as different point spread functions and different T1 weighting. Z-scores of [NAA] in patients with SIVD and patients with AD are listed in table 3 for various brain regions. In SIVD, cortical [NAA] was lower in the frontal and parietal lobes compared to the medial temporal lobe (F = 3.4, df = 2,113, p < 0.05). In contrast to SIVD, patients with AD had lower cortical [NAA] in the medial temporal lobe than in the frontal and parietal lobes (F = 4.6, df = 2,113, p < 0.02). For comparison, z-scores of MRI derived volumes of hippocampus and cortical gray matter in these patients are also listed in table 3.

Table 3

[NAA] z-scores of medial temporal lobe and parietal and frontal lobe gray matter in SIVD and AD and z-scores of MRI-derived volumes for comparison

Z-scores	SIVD	AD
[NAA]
Medial temporal lobe*	−0.4 ± 0.8	−1.0 ± 0.6†
Frontal lobe gray matter	−1.1 ± 0.8‡	−0.5 ± 0.8
Parietal lobe gray matter	−1.7 ± 0.8‡	−0.9 ± 0.9
Volume§
Hippocampus	−1.2 ± 0.9	−2.1 ± 1.2
Cortical gray matter	−1.8 ± 1.2	−1.4 ± 0.9

Open in a separate window

^*Includes hippocampus.

^†p < 0.02, Comparing institutional units AD [NAA] of medial temporal lobe to frontal or parietal lobes.

^‡p < 0.05, Comparing institutional units SIVD [NAA] of frontal or parietal lobes to medial temporal lobe.

^§MRI-derived volumes.

SIVD = subcortical ischemic vascular dementia; AD = Alzheimer’s disease.

According to our hypothesis that subcortical vascular disease would cause damage to cortical neurons by disrupting subcortical-cortical connections, we expected that cortical [NAA] losses should be related to the extent of WMH and subcortical infarctions, independent of cognitive status. As predicted, cortical [NAA] losses in patients with SIVD, averaged across the frontal and parietal lobe, were correlated with increased WMH volumes (r = −0.54, p < 0.02) and lacune numbers (r = −0.59, p < 0.02). Similarly, cortical [NAA] losses in CN subjects were also correlated with increased WMH volumes (r = −0.31, p > 0.02) and lacune numbers (r = −0.65, p < 0.02). Figure 1 illustrates the relationships between cortical [NAA] loss and WMH or lacunes, separately for patients with SIVD and CN subjects. Patients with AD, in contrast, showed no correlation between cortical [NAA] and WMH (r = −0.22, p > 0.05). However, power to detect a correlation of about 0.2 was not above chance with this number of patients. Because patients with AD had fewer than two lacunes by definition, correlations between cortical [NAA] and lacunes were not tested for AD. To explore further the idea that lacunar infarctions in strategic subcortical regions induce damage to remote cortical neurons, we tested whether [NAA] loss in the frontal cortex would be related to thalamic lacunes. This analysis was performed with data from SIVD, AD, and CN groups pooled. Using presence or absence of thalamic lacunes as criterion to dichotomize subjects with at least one lacune and accounting for MMSE differences between the groups, we found that thalamic lacunes were associated with 16% lower [NAA] (F = 6.7, df = 2,31, p < 0.02) in the frontal cortex but not in the parietal cortex (F = 2.2, df = 2,31, p > 0.1) compared to lacunes outside the thalamus. The association between cortical [NAA] and thalamic lacunes is shown in figure 2.

Open in a separate window

Figure 1

Relationship between cortical [NAA] and subcortical vascular disease, assessed by white matter hyperintensities (WMH) and number of lacunes in patients with subcortical ischemic vascular disease (●) and cognitively normal subjects (○). WMH volume was indexed to percent of normal white matter and [NAA] is represented in arbitrary units (au).

Open in a separate window

Figure 2

Cortical [NAA] of frontal and parietal lobe as a function of presence (+Thal) or absence (−Thal) of thalamic lacunes in subjects with subcortical infarctions. Also indicated are mean values and 1.0 standard deviations for each subgroup. Data from patients with subcortical ischemic vascular disease (), patients with Alzheimer disease (▲), and cognitively normal subjects (●) with lacunes are pooled.

Finally, we tested the extent to which [NAA] measurements added to MRI volume measurements improve differentiation between SIVD and AD. Results are summarized in table 4. Using logistic regression to predict diagnosis, we first entered hippocampal volume into the regression model, because the hippocampus yielded the largest difference of z-scores between patients with SIVD and patients with AD (z = −1.2 in table 3). The contribution of hippocampal volume in predicting diagnosis was substantial (p = 0.005), achieving an overall classification of 79% with specificity for SIVD of 92%, but sensitivity of only 36%. The area under the ROC curve was 0.83. Adding [NAA] of parietal cortex, which had the second largest z-score difference (−1.7 vs −0.9), improved overall correct classification from 79% to 89% (p = 0.02). This was primarily due to an increase for SIVD in sensitivity from 36% to 73% and a slight increase in specificity from 92% to 94%. Consequently, the area under the ROC curve also increased from 0.83 to 0.92. The scatter plot in figure 3 depicts the separation between patients with SIVD and patients with AD with hippocampal volume and [NAA] of parietal cortex. Adding more measures, such as WMH volume, gray matter volume, or frontal lobe cortical [NAA], yielded no further significant improvement. To test separation of patients with SIVD from CN subjects, we first entered cortical gray matter volumes from MRI into the regression model, because this measure had the largest z-score for SIVD (−1.8 according to table 3). The contribution from gray matter volume in predicting group was significant and yielded for SIVD an overall correct classification of 88% with 55% sensitivity and 96% specificity and an area under the ROC curve of 0.88. Adding [NAA] of parietal cortex or other measures did not further improve classification (all p > 0.2).

Open in a separate window

Figure 3

Discrimination between patients with subcortical ischemic vascular disease (●) and Alzheimer disease (○) by MRI-derived hippocampal volume and magnetic resonance spectroscopic imaging–derived cortical [NAA] of parietal lobe.

Discussion

[NAA] reductions predominantly in frontal and parietal gray matter in patients with SIVD, while sparing white matter and medial temporal lobe, are consistent with findings from several previous MRSI reports on SIVD.^21–24 However, previous MRSI studies sampled data from only a few, manually selected brain regions and were confounded by gray/white matter partial volumes, potentially introducing a sampling bias. The current study, utilizing all ¹H MRSI voxels in brain slices for analysis and regressing signal variations against voxel tissue composition, obtained metabolite intensities differentially for “pure” gray matter and white matter, greatly reducing complications from partial volume sampling. Therefore, findings from this study establish more firmly than before that cortical [NAA] reductions in SIVD were independent of morphologic brain changes and, furthermore, that regional [NAA] differences between SIVD and AD are not simply a sampling artifact. Although some [NAA] losses in the temporal and parietal cortex in patients with SIVD could be explained by concomitant AD, this does not explain [NAA] losses in the frontal cortex in SIVD, because AD was not associated with [NAA] losses in the frontal brain. Another important finding in SIVD was that [NAA] losses in cortical regions correlated with subcortical infarction load and WMH, implying that subcortical vascular disease was responsible for cortical changes in SIVD. There are several possible explanations for this finding. First, and most likely, is that subcortical infarctions cause functional deafferentation of the cerebral cortex, sometimes termed subcortical-cortical diaschisis. This is consistent with PET studies showing in SIVD hypometabolism and hypoperfusion in cortical regions,^25,26 especially in the frontal lobe.²⁷ In support of this view, MRS studies on animals have shown trans-synaptic decrease of NAA levels following acute deafferentation without neuronal loss.²⁸ It is therefore conceivable that the cortical [NAA] losses in SIVD could indicate deafferented neurons in a state of functional inactivity with a possibility for recovery rather than frank neuron loss. A second explanation is that neurons are damaged or lost via transneuronal degeneration, secondary to subcortical infarctions.²⁹ With the assumption that [NAA] reflects neuron density, the results imply further that a secondary degeneration causes disproportionately greater loss of neuronal than non-neuronal cells.

A third possibility is that cortical [NAA] reductions are due to cortical ischemia, with or without microinfarctions in the cortex, undetectable with MRI. There are two arguments against this view. First, quite remarkably, we found no [NAA] reduction in white matter in SIVD (or in AD). It would seem reasonable that a widespread ischemic process, which affected cortex and caused subcortical infarctions and WMH, would also produce [NAA] reduction in white matter. Second, [NAA] reductions can occur preferentially in the frontal lobe, as the comparison between subjects with and without thalamic lacunes showed. In presence of a generalized ischemic process, however, one would expect that the entire cortex would be affected, not that the frontal region would be singled out. Eventually, it will be necessary to obtain autopsy information to exclude microinfarctions and concurrent AD as potential cause of cortical [NAA] reduction in SIVD.

The finding of unchanged [NAA] in normal-appearing white matter in SIVD, while [NAA] was reduced in the cortex and regions of WMH, is surprising because the predominant disease process in these patients is thought to be subcortical. In another quantitative MRSI study of SIVD,²⁴ which used analysis of a few voxels of interest, we reported reduced [NAA] of white matter. However, this result was not independent of spurious contributions from [NAA] of WMH in SIVD, which is difficult to correct if only a small number of voxels is sampled. Our finding that the addition of parietal [NAA] to MRI measurements of hippocampal volume improved differentiation between SIVD and AD may be of clinical importance as treatments for the two disorders will be likely different. However, an overlap between the groups remained whether cortical [NAA] and hippocampal volume were used individually or together. Therefore, this will not likely lead to individual patient diagnosis, unless more information is incorporated, such as MRS measurements of myo-inositol³⁰ or topologic measurements of brain atrophy.^31–33

A major limitation of this study is that we used clinical and not autopsy data for patient classification, because it is generally recognized that an accurate prediction whether a patient has AD or SIVD is difficult to make antemortem. So far, autopsy data are available from only five patients with AD and three control subjects, but not yet from patients with SIVD. Two of the five autopsied patients were confirmed AD cases, whereas one had diffuse Lewy body disease and two had mixed AD/SIVD pathology, indicating the complications with accurate antemortem diagnoses. Thus, some patients with SIVD in this study may have had concurrent AD pathology, and some of the patients with AD likely had vascular disease and may have lacked AD pathology. Nevertheless, the finding of a different regional pattern of [NAA] loss in SIVD compared to AD is emphasized by the possibility of such misclassifications, because the ability to distinguish patterns between groups is reduced by misdiagnosis of subjects. There are several other limitations. First, a few patients with AD were taking cholinesterase inhibitors and anticholinergic drugs that could have altered brain metabolite levels, potentially enhancing metabolite differences between AD and SIVD. Secondly, relaxation measurements could not be performed owing to prohibitively long acquisition times. Therefore, [NAA] reduction in SIVD could be due in part to T2/T1 changes rather than to reduced concentration. Finally, detection of cortical atrophy and lacunes was compromised, because tissue segmentation relied on T2-weighted MR images with relatively low resolution (4.2 mm³). It is possible that different results would be obtained if tissue segmentation was performed with higher resolution.

Acknowledgments

Footnotes

References