Effects of Subcortical Cerebral Infarction on Cortical Glucose Metabolism and Cognitive Function FREE

CEREBRAL infarction as a cause of dementia has been well established for many years. Tomlinson and colleagues¹ clearly differentiated dementia caused by cerebral infarction from Alzheimer disease (AD) and applied the term arteriosclerotic dementia to cases of dementia with extensive infarction and minimal AD-type change. Hachinski et al² refined the concept by noting that cerebral infarction, rather than chronic ischemia, was the defining characteristic of this syndrome that they termed multi-infarct dementia. Within the past several years, new clinical criteria for the diagnosis of vascular causes of dementia have been suggested,^{3 - 4} and the term ischemic vascular dementia has been applied in a further attempt to delineate the different physiological causes of dementia due to cerebrovascular disease.

Despite these considerable advances in nosology, much uncertainty concerning the pathophysiology and clinical significance of vascular dementia persists. Estimates of the frequency of the disorder vary considerably,^{5 - 6} and diagnostic confusion extends even to neuropathological circumstances in which cerebrovascular disease and AD pathology coexist.^{7 - 8} These difficulties are related to problems with both clinical and pathological diagnosis and to the considerable heterogeneity of vascular dementia.⁹

Functional neuroimaging can potentially inform the discussion on vascular dementia by better defining pathophysiological mechanisms. In AD, reductions of regional cerebral metabolic rates of glucose (CMRglc) predominate in temporal and parietal cortex.^{10 - 12} Positron emission tomography (PET) can also define cortical hypofunction at sites remote from cerebral infarction and thus can potentially provide a more complete picture of the effects of cerebrovascular disease on the brain than structural imaging. For these reasons we chose to use PET and the glucose metabolic tracer fludeoxyglucose F 18 to determine relationships between glucose metabolism and cognitive impairment in patients with subcortical infarction. Only patients with lacunar cerebral infarction in subcortical structures were studied to reduce the heterogeneity of the disorder. Furthermore, the study design eliminated problems associated with different diagnostic criteria for vascular dementia by evaluating only relationships between cognitive function, subcortical lacunar infarction, and cortical metabolism.

Subjects were recruited from a university dementia clinic and a Department of Veterans Affairs system of hospitals and clinics in northern California. Subjects were identified by review of computed tomographic (CT) or magnetic resonance imaging (MRI) scans and invited to participate if they had 1 or more subcortical lacunar infarcts and met entry criteria. Eligible and willing subjects subsequently underwent a thorough evaluation, consisting of a medical and neurologic evaluation and laboratory screening to exclude reversible dementias. Comprehensive neuropsychological testing included the Mini-Mental State Examination (MMSE)¹³ as a measure of global cognitive function. In addition, all subjects had a research MRI scan performed. Subjects were excluded if they had cortical infarction, or if they had evidence of diseases or conditions that could affect cognitive function.

For the purposes of this study, subjects were defined clinically as "cognitively normal (CN)," "cognitively impaired (CI)," or "with dementia (D)." The diagnosis of dementia was arrived at by taking into account both medical history and neuropsychological testing and the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV)¹⁴ criteria were used. Subjects were judged to be cognitively impaired if they scored 0.5 on the Clinical Dementia Rating Scale¹⁵ and showed mild cognitive decline on neuropsychological testing.

In addition to cognitive classification, subjects were categorized as having lacunae or not by a single neuroradiologist (D.N.), who was blinded to the cognitive status of the patients. All ratings of stroke were performed on MRIs obtained on a single scanner using the same imaging protocol and the same criteria for definition of a lacune.

The combination of 3 cognitive status groups times 2 stroke groups (stroke and no stroke) yielded 6 potential groups. Because this report is concerned with vascular disease, we describe the results in 3 stroke groups: subcortical stroke and normal cognitive function (S-CN), subcortical stroke and cognitive impairment (S-CI) and no dementia, and subcortical stroke and dementia (S-D); and the control group (the group without stroke or dementia).

Magnetic resonance imaging was performed on a Siemens 1.5-T Magnetom VISION (Siemens, Erlangen, Germany) system equipped with a standard quadrature head coil. The imaging protocol included a sagittal T₁-weighted localizer scan, oblique axial double-spin echo scans angled parallel to the optic nerve, and a volumetric 3-dimensional magnetization prepared rapid gradient echo (MP-RAGE)–T₁ data set angled perpendicular to the double-spin echo image planes. The double-spin echo parameters were repeat time, 2500 milliseconds; first echo time, 20 milliseconds; second echo time, 80 milliseconds; 0.94-mm² resolution; and 48 to 51 contiguous 3-mm-thick slices covering the entire brain. The MP-RAGE parameters were repeat time, 10 milliseconds; delay time, 250 milliseconds; echo time, 4 milliseconds; flip angle, 15°; 1.0-mm² resolution; and 1.4-mm-thick partitions.

Fludeoxyglucose F 18–PET studies were performed on a CTI/Siemens ECAT EXACT HR PET scanner¹⁶ within 3 months of clinical evaluation and MRI. A subset of subjects had quantitative studies using a radial artery catheter to collect a blood input function.¹⁷ All subjects performed a continuous recognition memory task during tracer uptake¹⁸ following injection of approximately 370 MBq of fludeoxyglucose F 18. PET scanning commenced approximately 40 minutes after tracer injection, with 40 minutes of emission data acquisition followed by a 20-minute transmission scan.

The PET data were analyzed using procedures developed locally for the analysis of whole brain PET data sets that are described in a separate publication.¹⁹ The approach uses 3-dimensional MRI data (MP-RAGE in this case) for anatomic-region specification and analysis of the PET data. Two-dimensional MRI slices are used to define regions of interest that are subsequently tiled together to form volumes of interest (VOIs). By coregistration of PET and MRI data sets,²⁰ PET counts are extracted from these VOIs and used with the arterial input function to calculate regional CMRglc.^{21 - 22} In a final step, the VOI metabolic rates (or the count values) are adjusted for the effects of partial volume caused by cerebral atrophy using the segmented MRI data set as prior information in the manner developed by Meltzer et al.²³

All VOIs were drawn by a single operator (L.T.K.) blinded to patient classification, using a set of rules for guiding the boundaries of the VOIs, which are shown in Figure 1. Regions studied included temporal cortex (superior and middle temporal gyri), dorsolateral frontal cortex, orbitofrontal cortex, inferior parietal lobe (supramarginal and angular gyri), occipital (calcarine) cortex, and hippocampus (subiculum, Ammon horn, dentate gyrus, and associated white matter tracts). Whole brain glucose metabolism was calculated by defining a VOI outlining the entire brain (both hemispheres, including posterior fossa and subcortical structures) and using the quantification and atrophy correction procedures described above to determine CMRglc. In subjects without arterial catheterization, the atrophy corrected whole brain counts were used to normalize regional atrophy corrected counts to determine relative brain metabolism. Whole brain was chosen for the normalization region because it permits regional comparisons by scaling according to global metabolism, without introducing the bias that could distort ratios when small brain regions are used for normalization.

Two separate sets of variables were available for analysis. In 18 subjects, arterial blood input functions were available, and the atrophy-corrected regional CMRglc was calculated for each VOI. These values were analyzed with a repeated-measures analysis of variance (ANOVA) using group-by-regions design. In the entire group of 32 subjects, the normalized regional CMRglc ratio was also evaluated with a repeated-measures ANOVA. When the repeated-measures ANOVA was significant, follow-up 1-way ANOVAs were performed, and all such tests were corrected for multiple comparisons using the Bonferroni correction. Because groups were not matched for education, significant results were followed up with analyses that used education as a covariate. Inclusion of this covariate did not affect the results obtained.

Table 1 shows the characteristics of the entire sample and the sample with arterial catheters and absolute regional CMRglc determination. There were significant differences in education across the groups, with subjects having cognitive impairments and dementia, in general, showing lower education than subjects with no cognitive impairments. Differences in total stroke numbers were significant only in the entire sample and only when control subjects with no strokes were compared with the other groups. There were no statistically significant differences in the regional distributions of lacunae across the 3 stroke groups. In particular, thalamic lacunae were no more common in subjects with dementia or cognitive impairments than in those with normal cognitive function, although the sample sizes in each subgroup were small.

Whole brain absolute CMRs of glucose are shown in Figure 2. Patients with S-D had the lowest CMRglc, which was only slightly lower than that of patients with cognitive impairments, while both groups were lower than the 2 groups with no cognitive impairments. The results of the ANOVA were significant (P=.05), with post hoc Fisher protected least significant difference tests revealing that patients with S-D had significantly (P<.05) lower CMRglc than both control and S-CN patients, who did not differ from one another.

Absolute regional CMRs of glucose are shown in Figure 3 for the 4 groups. The 2 groups with no cognitive impairments show very similar patterns, which are almost superimposed and clearly higher than the 2 groups with cognitive impairments in all regions but the calcarine cortex. The ANOVA approached significance for this comparison (P=.08). When we combined the control and S-CN into one group and the S-CI and S-D group into another group, the repeated-measures ANOVA showed a highly significant (P=.008) group effect, along with a significant region effect (P<.001) and no interaction. This indicated that the cognitively normal subjects showed higher regional CMRglc than the subjects with cognitive impairments across multiple brain regions, consistent with the findings for whole brain CMRglc and Figure 3.

Results of the metabolic ratios are shown in Figure 4. Normalization of regional CMRglc eliminates the scaling effect of global CMRglc but reveals differences in the patterns in regional CMRglc. This is demonstrated by the ANOVA results that showed a borderline group effect (P=.09), a significant region effect (P<.001), and a significant interaction (P=.02). A series of ANOVAs for each brain region demonstrated that the right dorsolateral frontal cortex was significantly different (P<.001) between groups, with Fisher post hoc tests showing that all 3 stroke groups had significantly (P<.05) lower regional CMRglc ratios than the controls. The group effect for right dorsolateral frontal cortex remains significant when a Bonferroni correction for multiple comparisons (12) is applied using a setwise α of .05.

To evaluate the relationship between cerebral infarction, cognitive function, and glucose metabolism, we performed 3 correlations. For the entire sample, there was no significant correlation between the number of strokes and performance on the MMSE (r=0.14; P=.46). For the group of 18 subjects who had absolute CMRglc determined, there was a higher, though still nonsignificant correlation between the number of strokes and whole brain CMRglc (r=0.36; P=.15). When whole brain CMRglc was correlated with the scores on the MMSE for all 18 subjects with absolute CMRglc determined, this correlation was significant (r=0.50; P=.03) (Figure 5). Removal of the 6 control subjects resulted in decrease of the correlation coefficient to 0.42, which was no longer significant (P=.17).

This study reveals a number of findings concerning the relationship between subcortical strokes and cognitive impairment. First, cognitive function is related to the overall degree of cortical glucose metabolism, with cognitively normal subjects showing higher metabolic rates overall than subjects with cognitive impairments and dementia. Furthermore, the degree of whole brain glucose metabolism is related to the degree of cognitive impairment, while the number of strokes is only weakly related to both of these factors. Finally, there appears to be a regionally specific effect of subcortical strokes on right dorsolateral frontal glucose metabolism.

The relationship between cognitive function and hypometabolism that we found is consistent with previous reports. It is well established that subcortical cerebral infarction may produce cortical hypometabolism and hypoperfusion (the phenomenon of diaschisis), especially when infarcts are located in thalamus and white matter.^{24 - 26} However, most such studies have reported inconsistent relationships between subcortical lesions, cortical hypofunction, and behavioral deficits. Even fewer studies have related cortical physiology to behavior in patients with generalized cognitive decline, and none have specifically explored these relationships in patients with only subcortical infarcts. Mielke et al²⁷ found that the volume of hypometabolic regions was related to MMSE score in both patients with AD and vascular dementia, though there was no difference between the groups. This study was quite different from the present report, however, since the patients with vascular dementia had both subcortical and cortical lesions, and there was no stroke control group permitting the differentiation of the effects of stroke from those of dementia. Sultzer et al²⁸ also found modest relationships between the severity of subcortical white matter lesions and degree of global hypometabolism and behavioral changes in a PET study of patients with subcortical cerebrovascular dementia.

Several confounding factors that may be related to dementia in stroke are accounted for in this study. Cerebral atrophy has been suggested as an independent risk factor for dementia in patients who have had a stroke.^{29 - 31} Because the effects of atrophy were minimized in this study, metabolism, independent of brain structure, appears to be associated with cognitive failure. Advanced age has also been associated with the development of dementia in stroke patients in many studies, as has lower education in some studies.^{32 - 35} Our subject groups did not differ by age, and although the group with stroke and mild cognitive impairment was somewhat younger, this was not significant. Furthermore, this slight age difference is unlikely to bias the results since age has relatively small, if any, effects on glucose metabolism over small age ranges^{36 - 37} and would result in lower metabolic rates in younger subjects, contrary to our findings in the S-CI group. It is also unlikely that education differences affected the results, since statistical correction for group differences did not change the findings.

It is interesting that the cognitive status of subjects appeared to be better related to cortical glucose metabolism than to the number of subcortical strokes. The number of strokes did not differ significantly between the groups and there were not strong correlations between MMSE and the number of strokes. This finding is congruent with current conceptualizations of behavior that place the cerebral cortex in the midst of a number of circuits involving subcortical connections that mediate frontal lobe,^{38 - 39} mnemic,^{40 - 41} and attentional⁴² functions. Thus, multiple subcortical lesions may cumulatively and synergistically⁴³ affect behavior through effects on cerebral cortex. However, for a variety of reasons, the simple number of lacunae may not be a determining factor in affecting these cortical circuits. For example, white matter signal hyperintensities (WMSHs), which we did not quantitate, may also exert independent effects on cortical function.^{28 ,44 - 45} Subjects in this sample had a wide range of WMSH severities that could have affected cortical CMRglc. As DeCarli et al⁴⁶ have reported that individuals with such hyperintensities have reduced frontal lobe metabolism, these WMSHs in our subjects could have contributed significantly to the cortical changes, obscuring the role of lacunae. Furthermore, the criteria we used for lacunae might have missed small subcortical infarcts that could have functionally significant effects. Finally, simply attempting to count lacunae is probably an overly simplistic way to detect their effects on cognition, since these effects are likely related to size and location, with some lesions more likely to have functionally significant effects than others. Nevertheless, as a first approximation, simple lacune number does not appear to be as strongly related to cognitive function as does cortical glucose metabolism.

The specific effects of subcortical lesions on right frontal lobe function may have several explanations. Subcortical lesions are well known to produce frontal lobe–type cognitive deficits,^{39 ,47} and subcortical vascular dementia may be a particularly likely candidate to produce such frontal-subcortical syndromes.^{43 ,48} Frontal reductions in regional cerebral blood flow, oxygen metabolism, and glucose metabolism have been seen in PET studies of patients with subcortical cerebrovascular disease.^{28 ,44} In addition, the cognitive task performed during this PET study places demands on working memory⁴⁹ that might activate prefrontal cortex.^{50 - 52} Regardless of the precise reason for these frontal lobe findings, it seems unlikely that a specific effect on the right frontal lobe is a necessary characteristic of this syndrome.

The global and regional effects differed by groups, with subjects having dementia showing lower global CMRglc than subjects with S-CN and controls, but all stroke subjects showing lower regional CMRglc ratios in the right frontal lobes than did controls. This suggests that global CMRglc is closely related to cognitive state, while regional CMRglc is related to the presence of strokes. It is tempting to speculate that this regional effect may be an early feature of subcortical infarction that only results in cognitive impairment after a global reduction of metabolism has been reached. This hypothesis could be tested with longitudinal observation of stroke subjects.

This study is not a definitive report of the pattern of glucose metabolic changes in patients with vascular dementia for several reasons. First, this study was designed as an assessment of the effects of subcortical lesions on cortical metabolism and cognitive function, and current diagnostic classification schemes for vascular dementia were not applied. Second, vascular dementia is heterogeneous, and is unlikely to present with a single specific pattern of hypofunction because the distribution of infarct location is variable. Multifocal patterns of hypometabolism and hypoperfusion have frequently been reported in this condition⁵³ and proposed to be diagnostic, but many studies report that there is no characteristic pattern of hypofunction in vascular dementia.^{54 - 55} Third, temporal and parietal hypometabolism and hypoperfusion, similar to that seen in AD, have also been reported in vascular dementia,^{28 ,56} which could be due either to the location of infarcts or to the coexistence of AD. In this study, the pattern of reduced glucose metabolism seen in patients with dementia was not that of temporoparietal hypometabolism. In fact, in the entire stroke group, this pattern of temporoparietal hypometabolism was only seen on visual inspection in 1 subject. While we do not propose a charactertistic pattern of hypometabolism in vascular dementia as an interpretation of our data, it is interesting to note that temporal and parietal hypometabolism were not specifically seen in subjects with dementia, providing some evidence that AD is not invariably the explanation of cognitive decline in subjects with stroke. However, DeCarli et al⁵⁷ have suggested that subcortical cerebrovascular disease in patients with AD may make the predominant temporoparietal AD pattern less distinctive by also depressing frontal lobe metabolism. These discussions must remain conjectural until neuropathological verification of the cause of dementia is obtained.

While the mechanism underlying dementia in patients with subcortical cerebrovascular disease is undoubtedly complex, this study clearly demonstrates that patients with subcortical infarction who are cognitively intact have higher metabolic rates than those with cognitive dysfunction. In addition, subtle changes in frontal lobe function can be detected in such patients, even those with normal cognition. These studies support the notion that the production of dementia in subcortical vascular disease is related to both global and regional effects on cortical metabolism that are in turn related to the clinical presentation of the patient.

Accepted for publication October 23, 1998.

This study was supported by the National Institute on Aging, Bethesda, Md, grants AG12435 and AG10129.

Reprints: William J. Jagust, MD, Department of Neurology, University of California, Davis Medical Center, 4860 Y St, Suite 3700, Sacramento, CA 95817 (e-mail: wjjagust@ucdavis.edu).