Association Between Severe Cerebral Amyloid Angiopathy and Cerebrovascular Lesions in Alzheimer Disease Is Not a Spurious One Attributable to Apolipoprotein E4 FREE

CEREBROVASCULAR lesions are fairly common in autopsied cases of Alzheimer disease (AD),^{1 - 3} but what role these superimposed vascular lesions (VLs) have in producing or exacerbating dementia remains controversial. Most autopsy studies have found that approximately 80% to 95% of the cases of AD have appreciable amyloid accumulations in the walls of the cerebral blood vessels, especially in small arterioles.^{4 - 6} We have previously reported an association between severe cerebral amyloid angiopathy (CAA) and cerebral infarctions in AD. The risk of infarction is particularly high in patients who have both hypertension and severe CAA.⁷ Patients with severe CAA have the highest relative risk for cortical microinfarcts and hemorrhages, but multiple lesion types are often present (eg, larger cortical infarcts and lacunes).⁸ The apolipoprotein E-ϵ4 (APOE4) allele, which is the major identified genetic risk factor for AD, is associated with increased β-amyloid deposition in AD-affected brains. The effects of APOE4 have been extended to cerebral blood vessels, where increased CAA is found in carriers, and both within AD and non-AD samples.^{9 - 11} However, APOE4 is also associated with premature coronary artery disease and atherosclerosis.^{12 - 15} Thus, it is plausible that the association between severe CAA and cerebrovascular lesions is a spurious one mediated by APOE4 and/or generalized atherosclerosis.

Therefore, we reviewed our autopsy series at the San Diego Alzheimer's Disease Research Center to test if severe CAA remains an independent risk factor for cerebrovascular lesions after controlling for the effects of APOE genotype. We hypothesized that severe CAA would be more strongly associated with VLs than would APOE4 genotype.

Cases included in this autopsy study met the following criteria: (1) neuropathological criteria for AD, according to both National Institute on Aging¹⁶ (based on age-adjusted senile plaque densities) and Consortium to Establish a Registry for Alzheimer's Disease criteria¹⁷ (definite or probable AD); and (2) Diagnostic and Statistical Manual of Mental Disorders, Revised Third Edition criteria for dementia.¹⁸ Most (n=247) had also undergone APOE genotyping. Most patients were followed up longitudinally at the San Diego Alzheimer Disease Research Center ambulatory clinic with the remainder evaluated in either a private practice or nursing home. All San Diego Alzheimer's Disease Research Center subjects underwent at least one standardized evaluation that included a medical history, physical examination, structured neurological examination, cognitive screening tests, blood tests, and a neuroimaging study.¹⁹ The history regarding hypertension was obtained from an informant (usually the spouse) by either a neurologist or a nurse as part of the medical history or the Hachinski Ischemia Scale.²⁰ The Hachinski Ischemia Scale was done at the initial visit to the San Diego Alzheimer's Disease Research Center and was repeated annually if any interval history was suggestive of stroke or transient ischemic attack. Hypertension was defined as a recurrent abnormal systolic blood pressure higher than 140 mm Hg or diastolic blood pressure higher than 90 mm Hg by history or direct measurement—taking antihypertensive medications was not required. History of coronary artery disease was defined as either prior myocardial infarction (ie, clinical diagnosis or by electrocardiogram) or a history of exertional angina. Blood samples for nonfasting serum cholesterol levels were drawn in the late morning. Cases with superimposed diffuse Lewy bodies on brain autopsy were included in this analysis. Informed consent for brain autopsy and APOE genotyping had been obtained for all cases.

Three hundred six cases (162 males and 144 females) met these criteria and had the neuropathological measures described below performed. Two subgroups of patients were created—those with severe amyloid angiopathy (SAA) and those with mild or absent amyloid angiopathy (MAA)—using, as we had previously, the midpoint (≥2.0) of the amyloid angiopathy (AA) severity scale (defined below) as the cutoff. The size and characteristics of these groups are summarized in Table 1. The group with MAA (n=248) included 76 cases with no appreciable CAA in the areas sampled. One hundred forty-five of the cases in this study were included in previous articles.^{7 - 8} We have not previously reported on the relationship between APOE genotype and VLs in our cohort.

Apolipoprotein E genotyping was performed on either brain autopsy or blood samples using a polymerase chain reaction–amplification technique adapted from Wenham et al²¹ with modifications as have been reported previously.^{22 - 23}

The details of our brain autopsy procedures have been published previously.^{7 ,11} All brains were divided sagittally. Following 10 days of formalin fixation, the left hemibrain was examined externally, serially sliced into 1-cm-thick coronal sections, and evidence of infarction noted. Tissue blocks were taken from all gross lesions, as well as from 13 other routinely sampled brain regions.⁷ Hematoxylin-eosin–stained preparations from all tissue blocks were examined. These sections sometimes revealed cortical microinfarcts that had escaped detection at the time of brain sectioning. The cerebral vessels in these sections were examined and cases with prominent intracerebral atherosclerosis, arteriosclerosis, or

arteriolosclerosis were noted. The frozen right hemibrain was examined separately and the presence of grossly detectable infarcts noted. After partial thawing, the right hemibrain was cut in 1-cm-thick sections and any apparent infarcts noted.

Neocortical senile plaques, both diffuse and neuritic, neurofibrillary tangles (NFTs), and AA were evaluated in 10-µm-thick, 1% thioflavin-S–stained sections using UV illumination (with a 440-µm band pass excitation filter). Entire neocortical sections were surveyed to find areas with the most lesions. Three ×125 microscopic fields (field size, 1.76 mm²) were counted for total senile plaques (TPs), and three ×500 microscopic fields (field size, 0.1 mm²) for NFTs. The results were then averaged to provide single TP and NFT counts for each brain region from each case. These single TP and NFT counts for each of 4 brain regions (ie, hippocampus, midfrontal cortex, inferior parietal cortex, and superior temporal cortex) were then averaged to provide overall TP and NFT scores. Separate neocortical senile plaque counts for neuritic plaques, which contained filamentous amyloid and swollen neurites, were determined in 217 cases and an overall neuritic plaques score was calculated analogously. Total senile plaque scores included both diffuse and neuritic forms.

The severity of AA was assessed semiquantitatively on thioflavin-S–stained preparations of hippocampus, midfrontal cortex, inferior parietal cortex, and superior temporal gyrus. A score of 0 meant that there was no thioflavin-S positivity in the leptomeningeal or superficial cortical blood vessels. A score of 1 reflected trace to scattered positivity in either the leptomeningeal or the cortical blood vessels. A score of 2 indicated that at least some vessels in the leptomeninges or neocortex had circumferential brightly staining amyloid deposits. A score of 3 corresponded to widespread circumferential thioflavin-S positivity in many leptomeningeal and superficial cortical vessels. A score of 4 meant that similarly SAA was combined with dysphoric changes, ie, thioflavin-S positivity emanating from severely amyloidotic blood vessels into surrounding neuropil. A single AA severity score for each case was calculated by averaging across brain regions. Of the 58 cases who met criteria for SAA, all but 1 case had at least 1 brain section with widespread circumferential vascular amyloid depositions (ie, score ≥3), and 52 cases had at least 2 such brain sections.

All of the AD-affected brains were classified as being with or without VLs. Any cases showing grossly detectable large infarcts, cortical microinfarcts, granular cortical atrophy, intracerebral hemorrhages, or lacunar infarcts were designated as AD with VLs. Acute, subacute, and old lesions were all included. Primary hemorrhages were rare in our cohort and, therefore, were not analyzed separately. Only 2 cases (both with MAA) had primary hemorrhages without infarcts. One case of SAA had a hemorrhagic infarct and 1 case of MAA had an infarction and a hemorrhagic lesion.

One-way analyses of variance were applied to continuous variables and χ² analyses were used for categorical variables. A multiple logistic regression analysis was performed to see which of the following independent variables (ie, AA severity, hypertension, age of death, number of APOE4 alleles, and sex), entered simultaneously, predicted the presence of cerebrovascular lesions. The above regression models were then modified, entering 1 additional independent variable at a time. These additional variables included dichotomous variables indicating if there was history of coronary artery disease, atrial fibrillation, or intracerebral atherosclerosis, and a continuous variable for serum cholesterol level. Two-tailed P≤.05 was considered statistically significant.

Comparisons between cases with vs without severe CAA showed that the group with SAA had significantly more cases with VLs than the group with MAA (32.7% vs 19% of cases, P=.02, χ² test). When only cases with CAA scores of 2.5 or higher were compared with the group with MAA, the result was essentially unchanged (33.3% vs 19%; P=.04, χ² test). However, when the cutoff was raised to 3.0 or higher, the intergroup difference did not reach statistical significance (32.1% vs 19%; P=.09, χ² test), probably owing to an insufficient sample (n=28) of cases with this severity of CAA. Twelve of the 19 cases with SAA and VLs also had multiple infarcts (eg, 5 had multiple lacunes and 5 had multiple cortical microinfarcts). Almost all of the individual infarcts were small (volume <10 mL, except for 3 infarcts between 10 and 50 mL).

The group with SAA also had a higher frequency of the APOE4 allele (55% vs 35%; P<.001, χ² test), and a modest, but statistically significant, increase in TP and NFT counts relative to the group with MAA (Table 2). The frequency of the APOE2 allele was also higher in the group with SAA relative to the group with MAA (7% vs 2%), with the group with MAA having less APOE2 than is usual in the general population.²⁴ The 2 groups had an almost identical mean age of death (79.8 years), and very similar mean age of onset and duration of illness (Table 1). There was an almost statistically significant increased proportion of males in the group with SAA (64% vs 50%; P=.07, χ² test). Available data on frequency of hypertension, coronary artery disease, and serum cholesterol levels showed these to be comparable in the groups with SAA and MAA (Table 1). The group with SAA had a greater frequency of atrial fibrillation (chronic or paroxysmal) than did the group with MAA.

To see if the increased cerebrovascular lesions in the cases with SAA could be related to their having somewhat more advanced intraparenchymal AD lesions, the neuropathological scores of cases with vs without VLs were compared within strata of CAA (ie, SAA and MAA). Within the cases of SAA, these analyses showed no significant intergroup differences, with a trend for cases with VLs to have slightly lower NFT counts (Table 3). Within the cases of MAA, those with VLs were older and had significantly lower NFT, TP, and neuritic plaques counts (Table 3).

When we examined CAA severity as a function of APOE genotype, we found the APOE4 homozygous cases had significantly elevated CAA scores (Figure 1). One-way analysis of variance showed the effect of APOE genotype on CAA severity was highly significant (P<.001, χ² test), and Sheffé-corrected t tests showed that the group with APOE4/4 was significantly different from both groups with APOE3/4 and APOE3/3. However, there was no statistically significant increase in VLs in association with the APOE4 allele—VLs were present in 24% of APOE4 homozygotes, 19% of APOE4 heterozygotes, and in 25% of the noncarriers. Across all cases with VLs, the overall APOE4 allele frequency was 36.6%, which was almost identical to that of the cases without VLs (39.0%). Removing cases with the APOE2 allele did not substantially alter these results.

Next, we stratified our sample by APOE genotype, and performed separate analyses on frequency of VLs within each genotype group (ie, APOE3/3, APOE3/4, and APOE4/4). This showed that within 2 (the APOE4/4 and the APOE3/3 groups) of the 3 strata, there was a statistically significant increase in VLs in association with SAA (Figure 2). For example, within APOE4/4 homozygotes, 47% of the cases of SAA had 1 or more VLs compared with 9.5% of the cases of MAA (P=.01, χ² test). Within the group with APOE3/4, there was no statistically significant difference in VLs between the 2 subgroups (27% in SAA vs 16% in MAA; P=.29, χ² test), although the trend was again for more VLs in cases of SAA. We also looked at the effect that hypertension may have on the frequency of VLs. We found that approximately 27% of the cases with AD who had hypertension also had VLs, which was statistically significantly greater than in the cases with AD who did not have hypertension (15.9% had VLs; P=.04, χ² test).

Our main logistic regression models showed that the presence of SAA (odds ratio [OR]=3.53; 95% confidence interval [CI]=1.49-7.39; P=.002) and advanced age of death (OR=1.07; 95% CI=1.01-1.13; P=.01, χ² test) were both significantly associated with VLs at autopsy. The effects of hypertension (OR=1.62; 95% CI=0.81-3.25; P=.17, χ² test), sex (P=.80, χ² test), and APOE genotype (P=.93, χ² test) were nonsignificant in this model. The regression models that included atrial fibrillation (OR=1.16; 95% CI=0.37-3.58), coronary artery disease (OR=1.27; 95% CI=0.56-2.88), or serum cholesterol level (OR=1.00; 95% CI=.99-1.01) showed that none of these variables had statistically significant effects. Adding these variables did not appreciably change the effects of SAA. When the Hachinski Ischemia Scale score (applying a cutoff of ≥7 vs ≤6), which is often used as a clinical predictor of vascular or "mixed" dementia, was added, it was a highly significant predictor (OR=5.34; 95% CI=1.66-17.39; P=.005, χ² test). Both the SAA and age effects remained significant (OR=3.24 and OR=1.07, respectively) in this model. When intracerebral atherosclerosis was added, it was the strongest predictor of VLs (OR=49.3; P<.001, χ² test), but did not appreciably alter the effects of SAA or age.

This study is the largest neuropathological series we are aware of in which the relationship between CAA, APOE genotype, and VLs has been examined. We show not only that severe CAA confers a greater risk of VLs in AD, but that this risk is elevated even within strata of APOE genotype. Therefore, the association between SAA and VLs in AD is not a spurious one owing to APOE4. Instead, it is very likely that severe CAA is etiologically related to the VLs observed. We have previously shown that patients with CAA are at particularly high risk for specific types of VLs, especially small cortical infarcts, microinfarcts, and cerebral hemorrhages.⁸ In this study, almost all of the VLs were infarctions rather than hemorrhages. Given that CAA predominantly affects leptomeningeal and small cortical arterioles, a causal relationship is particularly likely with the associated small infarcts in the neocortical ribbon.⁷ It has also been suggested that CAA may cause deep infarctions by causing stenosis of the meningocortical segments of long perforating arterioles and secondary hypoperfusion of "watershed" areas (eg, deep subcortical white and gray matter prone to infarction).²⁵ Prior work by others has also strongly implicated CAA in producing ischemic lesions,^{26 - 27} in addition to the better known association with intracerebral hemorrhage.²⁸ The relative scarcity of cerebral hemorrhages in AD, despite abundant CAA, may be related to their very low prevalence of APOE2 allele, which predisposes toward primary intracerebral hemorrhage to a greater extent than APOE4.²⁹ Mandybur³⁰ described a "cerebral amyloid angiopathy-associated vasculopathy" with fibrinoid necrosis, hyaline arteriolar degeneration, obliteration of the intima, and microaneurysms. Our cases of SAA all had widespread circumferential amyloid deposits in their cerebral vessels, making the obliterative changes associated with advanced CAA a likely cause of infarctions.

Although APOE4 has been well demonstrated to increase the risk of premature coronary artery disease and may increase generalized atherosclerosis,^{12 - 15} its role in stroke and vascular dementia is much less clear. Among patients with stroke, some have found a modest increase in carriers of APOE4³¹ while other studies have found negative³² or opposing³³ results. In studies of clinically defined vascular dementia, the results are also conflicting, with many,^{34 - 36} but not all,^{37 - 38} finding an overrepresentation of APOE4. In contrast, neuropathologically defined groups of "pure" vascular dementia cases have generally been found to have no elevation in APOE4 allele frequency, while mixed AD/vascular dementia cases have shown increased APOE4.^{10 ,39} In our cohort with AD, there was no association between specific APOE genotypes and the prevalence of VLs. In this regard, our cases of AD with the APOE4 gene do not seem to have a more "vasculopathic" subtype of AD. However, this does not mean that cerebrovascular factors are unimportant in producing or exacerbating AD dementia independently of APOE. While we previously reported an increase in vascular-related cause of death among APOE4 carriers, this association was owing to a significant increase in cardiac-related deaths, while there was only a weak trend toward an increase in cerebrovascular-related mortality.⁴⁰ Our results are somewhat in conflict with those of Premkumar et al,¹⁰ who reported a moderate increase in the APOE4 allele frequency among their cases of AD with VLs, but a statistically nonsignificant difference in the overall APOE genotype distribution. Differences in patient selection could account for the discrepancies in that our center primarily recruits patients who meet criteria for probable AD,⁴¹ rather than vascular or mixed dementia. Supporting this is the very high prevalence of VLs (36%) found by Premkumar et al¹⁰ in their autopsied cases of AD, compared with only 22% (66 of the 306 cases) in this study. Unfortunately, Premkumar et al did not report the strength of the association between CAA severity and VLs. Caution is also advised before generalizing our results to younger cohorts: APOE4 confers its risk for AD in a highly age-dependent fashion⁴² and could conceivably confer its risk for stroke similarly.

Our data on atherosclerosis are consistent with Masuda et al,⁴³ who found no systematic relationship between severity of atherosclerosis and AA. However, we did find an unexpected association between SAA and atrial fibrillation. Since this was based on relatively few cases with atrial fibrillation (n=18 total), independent replication of this finding in another clinicopathological series is necessary. It is conceivable that poor cerebral perfusion caused by atrial fibrillation could facilitate CAA, perhaps by increasing amyloid production by smooth muscle cells and/or neurons.^{44 - 45}

Although hypertension was not a significant predictor in our regression models, it probably plays an important role in mixed dementia. We previously reported an apparent synergism between severe CAA and hypertension in producing cerebral infarction in AD.⁷ When an interaction term (severe CAA×hypertension) was added to the regression models in our study, this interaction term was highly significant and all the other potential predictors except age were statistically nonsignificant (results not shown here, but see references 7 and 8). This shows that hypertension particularly predisposes to VLs in the cases of SAA cases and/or that patients with hypertension who develop SAA are the most likely to have VLs.

While we found severe CAA to be a strong predictor of VLs, APOE4 genotype was not. The mechanisms by which CAA produces VLs of various types need to be further elucidated, as these could allow more effective interventions for patients with mixed AD/vascular dementia, an entity nearly as common as the frequently cited "second most common cause of dementia"—dementia with Lewy bodies.^{46 - 47} Direct effects of β-amyloid species on the cerebrovasculature have been recently described. Low concentrations of β-amyloid can induce vasospasm,⁴⁸ which could be of physiological significance when platelets, the major source of circulating β-amyloid, come into proximity with endothelial cells.^{49 - 50} Van Nostrand et al⁵¹ showed that β-amyloid can produce smooth muscle cell degeneration in vitro. Also, an agonist effect of aggregated amyloid on tissue-type plasminogen activator has been described, which may help account for the predilection for cerebral hemorrhages.⁵² This implies that elderly stroke patients with CAA could be particularly prone to hemorrhage after tissue-type plasminogen activator infusion. While knowledge of APOE genotype may help estimate the risk of CAA, most of the variance in CAA severity, within our cohort, remains unexplained by APOE genotype alone.¹¹ It is important that we improve our ability to diagnose CAA during life, which is difficult at present either prior to the first cerebral hemorrhage or in the setting of dementia.

Accepted for publication February 11, 2000.

This investigation was supported by grant AG 05131 from the National Institute on Aging, National Institutes of Health, Bethesda, Md.

Presented in part at the Vascular Factors in Alzheimer's Disease meeting, Newcastle, England, May 26, 1999.

We are grateful to Kathy Foster and Richard DeTeresa for technical assistance, and to Mary Pay, Deborah Fontaine, and Brock Riggins for obtaining and/or reviewing the subjects' medical histories.

Corresponding author: John Olichney, MD, Neurology Service (127), Veterans Administration Medical Center, 3350 La Jolla Village Dr, San Diego, CA 92161.