Cholesterol and LDL Relate to Neuritic Plaques and to APOE4 Presence but Not to Neurofibrillary Tangles

Abstract

Elevated serum total cholesterol (TC) has been considered a risk factor for Alzheimer's disease (AD), but conflicting results have confused understanding of the relationships of serum lipids to the presence of AD in the elderly.

Methods

To clarify these issues, we evaluated correlations of admission TC, low-density (LDL) and high-density (HDL) cholesterol directly with the densities of Alzheimer hallmarks--neuritic plaques (NP) and neurofibrillary tangles (NFT)--in nursing home residents (n=281).

Results

Significant positive associations of TC and LDL with NP densities were found in both the neocortex (TC: r=0.151, p=0.013 and LDL: r=0.190, p=0.005) and the hippocampal/entorhinal (allocortical) region (TC: r=0.182, p=0.002 and LDL: r=0.203, p=0.003). Associations of HDL with NP were less strong but also significant. In contrast, after adjustment for confounders, no correlations of NFT with any lipid were significant.When subjects with any non-AD neuropathology (largely vascular) were excluded, the TC-plaque and LDL-plaque associations for the remaining “Pure AD” subgroup were consistently stronger than for the full sample. The TC- and LDL-plaque correlations were also stronger for the subgroup of 87 subjects with an APOE ε4 allele.

Conclusions

The findings indicate that serum TC and LDL levels clearly relate to densities of NP, but not to densities of NFT. The stronger associations found in the subgroup that excluded all subjects with non-AD neuropathology suggest that cerebrovascular involvement does not explain these lipid-plaque relationships. Since the associations of TC/LDL with NP were particularly stronger in ε4 carriers, varying prevalence of this allele may explain some discrepancies among prior studies.

INTRODUCTION

High serum total cholesterol (TC) levels have been proposed as a risk factor for dementia, but various studies in elderly populations have provided differing results. Since TC can be medically modified, demonstration of such a relationship would suggest a potential for beneficial intervention. A number of investigators have reported positive relationships of serum lipids with the clinical diagnosis of Alzheimer's disease (AD), dementia, or cognitive impairment [1–5]. Higher TC and low-density cholesterol (LDL) have also been associated with more rapid cognitive decline in elderly subjects with AD [5–7]. Others have not found such relationships [8, 9] and some have reported negative associations of these serum lipids with AD or dementia or cognitive decline [10–13]. The time of life when lipids are sampled (i.e., mid vs. late-life) may partially explain such varying results [14–16]. Other possible explanations include epidemiologic differences among the study populations, as well as problems in clinically establishing the presence of dementia in individuals and in distinguishing etiologies among the different dementias [17–26]. In order to avoid limitations of clinical methodology, a few investigators have employed neuropathological assessments to establish and identify dementing processes. However, epidemiologic variations among the study populations, and differences of diagnostic criteria and of neuropathological techniques employed have continued to impair inter-study comparability [27–31].

In an effort to bypass the various clinical and pathological diagnostic issues, this study examined the associations of serum lipids with the traditional neuropathological features of AD--neuritic plaques (NP) and neurofibrillary tangles (NFT)--in nursing home residents. To our knowledge, only one prior study directly examined these relationships [28]; however, that study dealt with a very different population than ours—community-dwelling rather than nursing home subjects, Asian-American rather than primarily Caucasian, all male rather than largely female, and with lower prevalences of dementia and of the ε4 allele of apolipoprotein E (APOE), the only well established gene associated with AD.

METHODS

Subjects and Procedures

All subjects were residents of the Jewish Home Lifecare (JHL), a nursing home academically affiliated with the Mount Sinai School of Medicine. Procedures were approved by the Institutional Review Boards of both institutions. JHL medical records provided demographic and medical information. Fasting TC or a more extensive serum lipid profile was obtained within a few days of admission. High-density lipoprotein (HDL) and triglyceride analyses—used to calculate low density lipoprotein (LDL)—were not performed in the early years of the study so the earliest subjects have TC values only. Analytic methods for serum lipids and for APOE genotyping have been described previously and were stable throughout the observation period [29, 30]. Serum TC and HDL analyses were calibrated at least quarterly. From mid-1986 to mid-2006 permission for autopsy was obtained for 456 JHL residents; 344 of these had complete information available on neuropathological measures, including standardized ratings of neuritic plaque and neurofibrillary tangle densities [32] in every brain area specified for study, as well as age at admission, age at death, TC values, APOE genotype, race, gender and medication use. Those residents who died within six months of admission (63 of 344) were considered to have had potentially unstable admission lipid values associated with late or terminal illnesses [33, 34] and were excluded from the study. The remaining 281 residents with complete information who died later than 6 months after admission were included in the analyses; 222 of these 281 residents had additional information on LDL and HDL values.

Neuropathological Assessment

For this correlative study, data were obtained from systematic neuropathological studies performed at the Brain Bank for Neurodegenerative Disorders of the Mount Sinai/J.J. Peters Bronx VA Medical Center. Neuropathologists were unaware of clinical information and of APOE genotype. The basic neuropathological procedures have been previously described in detail [35], and were based upon the brain examination protocol of the Consortium to Establish a Registry for Alzheimer's Disease (CERAD) [32] for selection of brain tissue samples, processing/neurohistological staining of paraffin sections, and assessment and (modified) quantification of AD-related lesions. The data for the present study consisted of the neuropathologic findings from four areas of the right neocortex: middle frontal gyrus (Brodmann area 9), superior-middle temporal gyrus (Brodmann area 21/22), inferior parietal lobule (Brodmann area 7), calcarine cortex (Brodmann area 17/18), and from two areas of the right allocortex: rostral plus caudal hippocampus, with adjacent entorhinal cortex. Paraffin sections were cut at 5 micron thickness for hematoxylin and eosin stain and at 8 micron thickness for modified Bielschowsky silver stain. Immunohistochemistry preparations were also carried out on 8 micron thick sections using antibodies directed against Aβ amyloid (Dako Corp., Carpinteria, CA), abnormal tau (TG3, a gift of Dr. P. Davies, Albert Einstein School of Medicine, Bronx) for neurofibrillary tangle formation and alpha-synuclein (Dako Corp., Carpinteria, CA) to label Lewy bodies. The presence and extent of AD-related changes were assessed in the sections stained with modified Bielschowsky's silver stain. The assessment included estimates of the grades of density for NP and for NFT as per CERAD protocol [32]. The density assessment was carried out using medium high power (×20) objective lens which gave the visual field magnification of ×300. Multiple (usually five) representative high-power fields were examined from each silver-stained section, and densities of both NP and NFT for each specific area were scored, using the standardized CERAD scoring method, as: 0 (absent), 1 (sparse), 3 (moderate) or 5 (severe/frequent). NP and NFT densities are presented as mean scores for the four noted areas of the neocortical region and for the two noted areas of the allocortical region.

Statistical Analysis

The associations of three serum lipid measures (TC, LDL and HDL) with two neuropathological averages (NP and NFT) were assessed in each of two brain regions (neocortex and allocortex) by simple Pearson's correlation, as well as by partial correlation to control for the associations of confounders with lipids and with neuropathology. The confounders used were: age at admission; age at death; gender; race (Caucasian versus other); current or prior use of a lipid-modifying agent; and presence of an APOE ε4 allele. Since AD-associated neuropathologic lesions and their relationships with clinical manifestations of dementia are not linearly related to age in the elderly [36, 37], both linear and quadratic effects of age at admission and age at death were included as confounders. We also evaluated whether the correlations between NP densities and lipid measures were significantly stronger than the same correlations between NFT densities and the lipid measures in each brain region, using the procedure of Meng [38], which takes account of correlations being measured in the same sample with a common variable.

We similarly analyzed two subgroups: (a) those subjects with uncomplicated AD neuropathology as defined by CERAD [32] and those with normal brains (“Pure AD” subgroup); and (b) all other subjects with any neuropathology other than AD (with or without complicating AD) (Complementary subgroup), again evaluating all these simple and partial correlations. Partial correlations of subgroups were compared using Fisher's z transformation.

In addition to controlling for the confounders by partial correlation, we further explored each of their interactions with serum lipids on NPs and NFTs, i.e., assessing whether the relationships of serum lipids with neuropathology differed according to the level of the confounder. Four of the confounders—sex, race, use of a lipid modifying agent, and presence of an APOE ε4 allele—were dichotomous. For the two subgroups of each dichotomous confounder, we calculated the simple and partial correlations of each serum lipid with each NP and NFT measure. In addition, for each dichotomous confounder, the difference between the associations of each lipid with each neuropathological measure for each of the two subgroups was assessed by stepwise regression, entering (a) all the confounders, including the dichotomy of interest; then (b) the lipid; and then (c) the interaction of the lipid with the dichotomous confounder. For the continuous variable confounders—age at admission and age at death—similar stepwise regression analyses assessed whether the relationships of lipids and plaques or tangles varied according to age in either a linear or quadratic pattern. We performed steps (a) and (b) as above, then entered the interaction of the lipid with linear age and, finally, the interaction of the lipid with quadratic age (i.e., age squared). For each confounder there were 12 interaction analyses (of 3 lipids with 2 neuropathological measures in 2 regions); we used the Bonferroni inequality to evaluate their overall significance at the 0.05 level. At least one interaction must be significant at the 0.004 level (0.05/12) to demonstrate that some association of a lipid with a neuropathology differs according to the level of the confounder.

RESULTS

The study sample was 75% female and 84% Caucasian (11% Black, 4% Hispanic, 1% Asian). Average age at admission was 83.1±8.0 (SD) yr; age at death 87.8±8.2 yr; and length of stay 4.6±3.8 yr. Only 16 % of residents were below age 80 at time of death. Mean ±SD values for serum lipids at admission were, in mg/dL (mmol/L): TC=201.7±51.4 (5.22±1.33), LDL=125.7±43.0 (3.25±1.11), and HDL=46.2 ±13.7 (1.19±0.35). Clinical dementia (defined as Clinical Dementia Rating ≥ 1.0 [39]) was present in 77% of these residents at admission, and increased to about 90% during the JHL stay. Eighty-seven residents (31%) had at least one APOE ε4 allele. During most of the period of observation, “statins” were rarely prescribed for the very old, so only sixteen residents (6%) had used or were using a “statin” or other lipid-modifying agent at admission. For the 281 subjects, the distribution of neuropathological diagnoses, using CERAD diagnostic criteria [32], was: normal brain, 37; only a potentially dementing process other than AD, 48; only uncomplicated AD neuropathology (definite, probable or possible), 101; and AD neuropathology plus a non-AD dementing process, 95.

Correlations of each serum lipid value with mean scores of NP and also of NFT for the neocortex (mean for the 4 areas examined) and the allocortex (mean for the 2 areas examined) are presented in Table 1 with no adjustments and also with adjustments for multiple confounders. In nearly all instances, values of partial correlation (i.e., adjusted for confounders) were the same or lower than those of the respective simple correlations, particularly so for TC and LDL; the results and comments below refer only to the partial correlations. Admission serum TC and LDL were each positively and significantly correlated with densities of NP in both neocortical and allocortical regions. Positive associations with regional NP densities were also significant for HDL, but less strong. In contrast, for NFT there were no significant correlations with TC, LDL, or HDL (although several were significant if not adjusted for confounders) (Table 1). We found that for TC and LDL, correlations with NP were significantly larger than those for NFT; for HDL, the correlations with NP and NFT did not differ significantly Fig. (1).

Table 1.

Correlations of Serum Lipids with Regional Plaque and Tangle Densities

	Total Cholesterol N=281				Low Density Cholesterol N=222				High Density Cholesterol N=222
Neuritic Plaques (NP)	Unadjusteda		Adjustedb		Unadjusteda		Adjustedb		Unadjusteda		Adjustedb
	rc	Pd	rc	Pd	rc	Pd	re	Pd	re	Pd	rc	Pd
Neocortexe	0.211	<0.001	0.151	0.013	0.250	0.001	0.190	0.005	0.156	0.020	0.145	0.034
Allocortexf (Hippocampal & Entorhinal)	0.215	<0.001	0.182	0.002	0.234	0.001	0.203	0.003	0.155	0.021	0.150	0.028
Neurofibrillary Tangles (NFT)
Neocortexe	0.136	0.022	0.049	0.420	0.154	0.021	0.081	0.236	0.137	0.041	0.118	0.084
Allocortexf (Hippocampal & Entorhinal)	0.104	0.081	0.007	0.911	0.108	0.108	0.041	0.555	0.084	0.214	0.044	0.518

^aPearson's correlations, without adjustment.

^bPartial correlations, adjusted for race, gender, presence of an APOE ε4 allele, use of a lipid-modifying agent, and linear and quadratic ages at admission and death.

^cr = correlation coefficient.

^dp = probability (two-sided).

^eMean of NP or NFT density ratings for four neocortical areas: midfrontal gyrus, superior temporal gyrus, inferior parietal lobule and calcarine cortex.

^fMean of NP or NFT density ratings for two allocortical areas: hippocampus and entorhinal cortex.

Fig. (1).

Correlations of Serum Lipids with Plaque and Tangle Densities. NP = density of neuritic plaques; NFT = density of neurofibrillary tangles. TC = total cholesterol; LDL = low-density cholesterol; HDL = high-density cholesterol. Neocortex = mean of NP or NFT density ratings for four neocortical areas, as noted. Allocortex = mean of NP or NFT density ratings for hippocampus and entorhinal cortex. Partial correlations are adjusted for gender, race, presence of an APOE ε4 allele, use of a lipid- modifying agent, and linear and quadratic ages at admission and death. *p ≤ 0.05, **p ≤ 0.01: significance of difference between NP and NFT correlations.

We also examined the relationships of lipids with NPs and NFTs for the subgroup comprised only of those subjects whose neuropathology was restricted to uncomplicated AD or to normal brain (Pure AD) and also for the Complementary subgroup of subjects with any non-AD neuropathology (with or without AD pathology). For the “Pure AD” subgroup, the relationships of TC and LDL with plaques were highly significant and consistently stronger than those of the full sample Fig (2). For the Complementary subgroup these TC and LDL relationships were consistently weaker than those of the full sample, and none was significant. The TC correlations with NP were significantly stronger for the “Pure AD” than for the Complementary subgroup in both the neocortex (p=0.04) and allocortex (p=0.03). In contrast, the weak associations of HDL with neocortical and allocortical NP that had been noted for the full sample were no longer significant for either the “Pure AD” or the Complementary subgroup Fig. (2). For the “Pure AD” subgroup, there were again no significant associations of any of the lipids with NFT, with the single exception for LDL in the neocortex (r=0.226, p=0.026). Similarly for the Complementary subgroup, none of the associations of any of these lipids with NFT were significant.

Fig. (2).

Correlations of Serum Lipids with Neuritic Plaque Densities. Full sample = all subjects: n = 281 for TC; n = 222 for LDL and HDL. Pure AD subgroup=only subjects with uncomplicated AD or with no pathology: n = 138 for TC; n = 106 for LDL and HDL. Complementary subgroup = only subjects with non-AD pathology (with or without AD pathology); n = 143 for TC; n = 116 for LDL and HDL. TC = total cholesterol; LDL = low-density cholesterol; HDL = high-density cholesterol. Neocortex = mean of NP density ratings for four neocortical areas, as noted. Allocortex = mean of NP density ratings for hippocampus and entorhinal cortex. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001: significance of partial correlation. Partial correlations are adjusted for gender, race, presence of an APOE ε4 allele, use of a lipid- modifying agent, and linear and quadratic ages at admission and death

Using the Bonferroni inequality criterion to evaluate interactions for each of the 8 confounders for the partial correlations, significant interactions were found only when distinguishing the 87 subjects with an APOE ε4 allele from the 194 subjects without this allele by their associations of lipids with NP or NFT. The partial correlations of each lipid and neuropathology for the ε4 positive and ε4 negative subgroups are presented in Table 2. Compared to the full sample, all associations were stronger for the ε4 positive subgroup and weaker for the ε4 negative subgroup. For the ε4 positive subgroup, correlations for TC and LDL with plaque densities were all significant (except one at p=0.054); and for the ε4 negative subgroup, none was significant despite the much larger sample size. The NP relationships were different for HDL than for TC and LDL: correlations with plaque densities for the ε4 positive and ε4 negative subgroups were all weak, differed little from each other, and achieved marginal significance only in the larger ε4 negative subgroup. We also analyzed the differences between correlations in the two subgroups for significance: the ε4 positive correlations with NP were significantly larger than the ε4 negative correlations for the allocortex for both TC (p=0.019) and LDL (p=0.002). Similar to the comparisons of “Pure AD” and Complementary subgroups, there were no significant correlations of NFT in any region with any of the serum lipids in either APOE subgroup, with the single exception of neocortical NFT with LDL in ε4 positive subjects (Table 2).

Table 2.

Partial Correlations of Serum Lipids with Regional Plaque and Tangle Densities by APOE ε4 Status^a

APOE ε4 Statusa n	Total Cholesterol				Low Density Cholesterol				High Density Cholesterol
	ε4 Positive 87		ε4 Negative 194		ε4 Positive 71		ε4 Negative 151		ε4 Positive 71		ε4 Negative 151
Neuritic Plaques (NP)	rb	Pc	rb	Pc	rb	Pc	rb	Pc	rb	Pc	rb	Pc
Neocortexd	0.216	0.054	0.133	0.069	0.337	0.007	0.147	0.079	0.166	0.190	0.163	0.051
Allocortexe (Hippocampal & Entorhinal)	0.413	<0.001	0.119	0.105	0.517	<0.001	0.107	0.200	0.083	0.514	0.187	0.025
Neurofibrillary Tangles (NFT)
Neocortexd	0.158	0.162	0.012	0.866	0.291	0.020	0.011	0.898	0.161	0.204	0.140	0.095
Allocortexe (Hippocampal & Entorhinal)	0.166	0.142	−0.029	0.696	0.219	0.083	0.004	0.958	0.122	0.338	0.046	0.584

^aε4 positive = subjects with one or two APOE ε4 alleles; ε4 negative = subjects absent an APOE ε4 allele.

^br = correlation coefficient. Pearson's partial correlations, adjusted for race, gender, use of a lipid-modifying agent, and linear and quadratric ages at admission and death.

^cp = probability (two-sided).

^dMean of NP or NFT density ratings for four neocortical areas: midfrontal gyrus, superior temporal gyrus, inferior parietal lobule and calcarine cortex.

^eMean of NP or NFT density ratings for two allocortical areas: hippocampus and entorhinal cortex.

DISCUSSION

The present study documents consistent relationships of serum lipid values with NP neuropathology but, after adjustment for confounders, not with NFT neuropathology. Specifically, TC, LDL, and HDL were all significantly associated with NP in the neocortical and allocortical regions generally considered to be relevant to AD. When these analyses were limited to only those subjects with pathologically uncomplicated AD or normal brain, the correlations of TC and LDL with plaque densities for this “Pure AD” subgroup were all noticeably stronger than those of the full sample, and all were highly significant. In contrast to TC and LDL, correlations for HDL-plaque density in this “pure” subgroup differed little from those of the full sample and, with the smaller sample size, were no longer significant. Therefore, the associations of NP with HDL are questionable. With a single exception, no significant association of NFT with any lipid was observed in the Pure AD or Complementary subgroups.

Comparisons with Prior Lipid-AD Studies

As noted, widely varying results have been reported in many clinically-based studies of serum lipid-AD associations in the elderly [1–5, 8, 10–12]. A few investigators employed neuropathological criteria to better identify the presence and type of the dementing condition [27, 28, 30, 31]. Among those reports, the brain areas studied, the pathological methods employed and the criteria for neuropathological diagnoses often differed. Our earlier reports [29, 30] and that of Kuo [27] dealt with very old subjects and followed CERAD criteria [32] for brain areas of study and for neuropathological definition of AD; each noted positive relationships of serum LDL and TC with AD. Pappolla, et al. [31], studying the hippocampus and adjacent temporal lobe, reported that serum cholesterol correlated significantly with amyloid deposition in subjects up to 55 years of age, but not at later ages.

In the present study, we sought to avoid limitations related to clinical dementia diagnosis or to neuropathological diagnosis by any specific set of criteria by associating serum lipid values directly with the densities of neuritic plaques and of neurofibrillary tangles in multiple, pertinent areas of the brain. One prior report, from the Honolulu-Asia Aging Study (HAAS) [28], also examined serum lipid values and NP and NFT densities in a population sample of comparable size and of similar ages. Those authors noted significant positive associations only for late-life serum HDL with neocortical and hippocampal NP and with neocortical NFTs. Our findings of significant associations of HDL with NP in the neocortical and allocortical (hippocampal/entorhinal) regions for the full sample were in agreement with those of the HAAS; and our observed association of HDL with neocortical NFT was similar but was short of significance. In contrast with their findings, we observed significant relationships of TC and LDL with neocortical and allocortical NP densities. The HAAS protocol [28] differed from ours in many important respects: it studied a community-based, all male, and all Asian-American population and also differed in design, procedures, choice of subjects for autopsy, statistical methods and prevalence of APOE ε4 genotype, any of which might have influenced results. As an instance, we found serum TC and LDL to be strongly related to NP densities in ε4 positive but not in ε4 negative subjects in both neocortical and allocortical regions (Table 2). If our sample had been similar to the HAAS in having had only 13% ε4 positive subjects—rather than our actual 31%—the lipid-plaque associations for LDL and TC would have been substantially weaker and more similar to those of the HAAS. In another neuropathological report of very old subjects with high prevalence of ε4 allele positivity (49%) [27], the results were in general agreement with our findings in that LDL and TC relate to the plaque-defined (CERAD) presence of AD. These comparisons illustrate how such differences in study population characteristics can lead to important differences in outcomes. It is quite possible that sample differences, such as the prevalence of ε4 positivity, can explain some major discrepancies not only between our results and those of the HAAS but also among the many studies noted above.

Lack of Concurrence of NP with NFT Densities

In this study, TC and LDL were found to be clearly related to NP but not notably related to NFT. For TC and LDL, the correlations with NP differed significantly from those with NFT in each brain region Fig (1). Although NP and NFT often occur together and are the recognized hallmarks of Alzheimer's disease, the relationships between them and the commonality between their neurobiological origins remain uncertain. In large, very old cohorts some brains have been found to contain only plaques or only tangles [40], and Giannakopoulos, et al. [41] did not find correlation of NP with NFT in the cerebral cortex. Associations of NP and NFT densities (or amyloid and tau content) with other variables have often diverged. Hyperglycemia and hyperinsulinemia were asociated with increased risk for NP but not NFT [42]. Several studies of subjects with pathologically documented AD (by NIA-Reagan and by CERAD criteria) found the presence of an APOE ε4 allele to be associated with NP burden in specific areas and overall, but not similarly associated with NFT density [43, 44]. Launer, et al. [28] found significant relationships of late-life serum HDL to NP in both the neocortex and the hippocampus, but to NFT only in the neocortex. It is common to find NFT but not NP in the hippocampus (and other allocortical areas) of cognitively normal elderly [25, 45–48]. In addition, the relationships of NP with NFT densities have been noted to differ with level of education [49] and also with age [36, 37, 47]. Since plaques appear to depend upon the accumulation of β-amyloid, and tangles upon the accumulation of tau protein, it is plausible that serum lipids may relate differently to the resulting neuropathological entities.

Effects of APOE Genotype on Lipid-Neuropathological Relationships

For the subgroup of subjects who carried the APOE ε4 genotype, all associations of TC and LDL with NP (but not NFT) were stronger than those of the full sample. Several other investigators have found the ε4 allele to be associated pathologically with an increased burden of NP, but not of NFT [43, 44], and recent imaging studies (with labelled PIB-PEF) have related ε4 presence to increased age-related cerebral amyloid deposition [50, 51] but not to tau levels in the spinal fluid of these individuals [51]. The ε4 allele has also been reported to be specifically related to plaque Aβ deposits [52, 53]. In contrast, others have found ε4 allele presence to be related to both NP and to NFT [54, 55].

Do TC and LDL have a Role in AD Pathogenesis?

The documentation of associations we observed does not, of course, identify a specific role for TC and LDL in AD pathogenesis. While the relationships of these serum lipoproteins with neocortical and allocortical plaque densities were all significant, the correlation coefficients for the full sample were only modest, indicating that factors other than serum lipids are also related to the density of NPs. As noted, we did find the TC-plaque and LDL-plaque correlations to be stronger in the subgroup that reflects uncomplicated AD, supporting an involvement of these serum lipids with AD. Others have sought additional support for the lipid-AD involvement by observing effects of manipulating these serum lipids on the course of AD. In recent publications, several investigators noted beneficial effects of lipid-lowering agents (primarily “statins”) in development of AD [56, 57], in slowing the rate of cognitive decline in already established AD [58, 59], and in reducing Alzheimer neuropathology [60]. In contrast, another study found no relation of statin use to AD-related neuropathology [61]; two large randomized, placebo-controlled clinical trials of lipid lowering and cardiovascular disease [62–64] and a recent smaller study [65] found no apparent effects of statin use on the incidence or progression of dementia. Similarly, Benito-Leon, et al. found no change of cognitive function in community-based elders [66].

Several cardiovascular risk factors have been associated with AD [67–72] and the association of coronary artery disease, APOE ε4 and AD was demonstrated directly in cases derived from our Brain Bank [73]. The role of vascular dysfunction in AD has recently been reviewed [74]. As serum cholesterol and LDL are such cardiovascular risk factors, it can be postulated that the lipid-NP associations found in the full sample resulted from contribution of cerebrovascular disease to the higher extent of NP. However, the present findings are inconsistent with such an hypothesis. The “Pure AD” subgroup (absent vascular neuropathology) had substantially higher plaque densities than the Complementary subgroup which had primarily vascular neuropathology (some with and some without associated AD) Fig. (2). It has been reported that the presence or absence of neuropathologically defined vascular lesions had no effect on the prevalence of AD markers [75], also suggesting independent development of vascular and AD lesions.

Our findings that HDL, as well as TC and LDL, was positively associated with plaque density might be considered “counterintuitive”, in view of their contrasting roles in cardiovascular disease. Although in the full sample HDL-NP correlations were similar to (but weaker than) those of TC and LDL (Table 1), in further analyses of subgroups these relationships remained consistent for both TC and LDL, but differed for HDL Fig. (2), (Table 2). Furthermore, this intuition would be appropriate only if the lipid-AD association were to depend upon some “vascular-related” process. To the contrary, elimination of subjects with vascular neuropathology strengthened the relationships of TC and LDL with NP, and did not appear to affect the relationships of HDL with NP Fig. (2). Some of the many “non-vascular” mechanisms by which serum lipoproteins could relate to the pathogenesis of AD and its neuropathology are discussed below.

There are suggestions from several disciplines that support a relationship of cholesterol with the development of AD. A number of genes related to brain and peripheral cholesterol or to lipoprotein homeostasis are associated with AD, although this need not reflect a common pathway [76–79]. Various findings have suggested mechanisms for the involvement of cholesterol homeostasis in the pathogenesis of AD [71, 72, 76–78, 80–87]. Among these, enzymatic activities responsible for Aβ production were inhibited by lowering of cellular cholesterol, and probably also by peripheral cholesterol depletion [82]. Kosicek and co-workers, using Niemann-Pick model cells, noted that cholesterol overload, as in N-P type C, led to increased amyloidogenic processing of amyloid precursor protein (APP) apparently taking place in lipid rafts, supporting a possible role of lipid rafts in the pathogenesis of Alzheimer's disease [88]. Significant cholesterol retention has been found in the AD brain and elevated cholesterol levels can increase (and cholesterol depletion can reduce) β- and γ-secretase activities and Aβ production [83]. A disruption of lipid rafts may favor the amyloidogenic APP processing pathway by bringing APP and β-secretase into close proximity and enabling the cleavage of APP to produce Aβ [87, 89]. Induction of hypercholesterolemia in animal model systems has led to inconsistent results [90]; however, some studies have shown these manipulations to promote AD-like neuropathology and cognitive dysfunction [86, 91–93]. It has recently been observed that increased mRNA and protein expression of the cholesterol transporter ABCA1 is highly correlated with severity of clinical dementia and also with densities of NP and NFT in AD hippocampus [87]. Similar results were observed for the master regulator of CNS cholesterol biosynthesis RXR/LXR [94]. Despite these observations, the relationships of serum cholesterol and brain cholesterol are not clear. While serum lipids do not normally cross the blood-brain barrier (BBB), they could affect CNS cholesterol homeostasis by acting on or through substances that can cross the BBB [77, 84, 95, 96], or by penetrating the BBB in abnormal circumstances [77, 78, 97–100]. Several reviewers have critically evaluated a broad range of in vitro and in vivo experimental findings in these areas [77, 101–104].

This study has several limitations. Our protocol was designed to evaluate the relationships of serum lipids specifically with the hallmarks of AD—plaques and tangles—and is not an analysis of serum lipid relationships with the full range of neuropathology of dementia. In the absence of adequate information on pre-nursing home admission serum lipid values or reliable estimates of pre-admission dementia duration, the temporal relationships of TC and LDL levels with the development of neuropathological abnormalities cannot be determined. Request for autopsy was routine practice and was granted for approximately 10% of all deaths. Although we were not able to identify specific sources of bias, some bias in granting of consent can affect results in any pathologically based study. Also, the findings should be generalized with caution, since most NH populations have characteristics that could affect the relationships of lipids to neuropathology, including advanced age, severity and frequency of comorbidities, high dementia prevalence, and dementia severity. The findings are strengthened by the sizeable sample and the well-defined pathological methods. In particular, by correlating serum lipid values directly with independently observed scores of plaque and tangle densities we avoided the potential problems in clinical definition of dementia and in pathological classification of dementing processes.