The Apolipoprotein E ϵ4 Allele and Decline in Different Cognitive Systems During a 6-Year Period FREE

POSSESSION OF 1 or more copies of the apolipoprotein E (APOE) ϵ4 allele is associated with an increased risk of Alzheimer disease (AD)^1- 2 but the mechanism underlying this association is unclear. Since a defining feature of AD is progressive loss of episodic memory, several researchers have hypothesized that the ϵ4 allele is selectively associated with episodic memory decline in older persons.^3- 5 Support for this hypothesis has been mixed, however. The ϵ4 allele has been associated with episodic memory impairment in some cross-sectional studies^3,6- 7 and with more global cognitive impairment in others.^8- 10 Although several longitudinal studies have examined the relationship of the ϵ4 allele to cognitive decline,^4,11- 20 few have assessed multiple cognitive systems^{4,13- 14,17- 19} and most of these have been based on only 2 observations during periods of 3 years or less and have conducted analyses on individual tests, which are subject to floor and ceiling effects. In addition, no previous longitudinal study has directly tested whether the association of ϵ4 with change in measures of episodic memory differs from its association with change in other cognitive measures.

We used data from the Religious Orders Study, an ongoing clinicopathological study of aging and AD in older Catholic clergy members, to investigate the differential effects of the APOE ϵ4 allele on change in episodic memory and other cognitive abilities. At baseline, participants were 65 years and older and free of clinical evidence of AD. They underwent annual clinical evaluations for up to 6 years, including detailed cognitive function testing and clinical classification of AD, with follow-up participation in survivors exceeding 95%. Composite measures of episodic memory, semantic memory, working memory, perceptual speed, and visuospatial ability, each based on 2 or more individual tests, were the main outcomes. In analyses, we first assessed the association of the ϵ4 allele with risk of developing AD. We then used a growth curve approach to estimate the association of ϵ4 with the baseline level of and annual rate of change in each cognitive system and to test whether the effects on any 1 system differed from the average effects on the other systems.

Subjects are participants in the Religious Orders Study, a longitudinal clinicopathological study of aging and AD in older Catholic nuns, priests, and brothers recruited from about 40 groups across the United States. Eligibility required age of 65 years or older, absence of a clinical diagnosis of dementia, and consent to annual clinical evaluations and to brain donation at death. The study was approved by the institutional review board of Rush-Presbyterian–St Luke's Medical Center (Chicago, Ill).

Participants enrolled in the Religious Orders Study by October 2000 were eligible for analyses in this study if they had a valid APOE genotype, did not meet criteria for dementia at baseline, and survived to the first follow-up date. Of 624 people who met these criteria, 611 (97.9%) completed at least 2 evaluations (mean, 5.5 evaluations; 97.3% of possible evaluations in survivors). Analyses are based on this group.

At baseline, each person had a uniform evaluation, which was repeated annually, with examiners blinded to previously collected information as previously described.^21- 24 It included a medical history, neurologic examination, cognitive function testing, and review of brain scan when available. On the basis of this evaluation, a board-certified neurologist classified people with respect to AD and other common neurologic disorders. The diagnosis of dementia and AD was based on the criteria of the joint working group of the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer's Disease and Related Disorders Association (NINCDS/ADRDA).²⁵ Dementia required a history of cognitive decline and impairment in at least 2 cognitive domains, one of which had to be memory for the dementia to meet AD criteria. Those who met these criteria and had another condition judged to contribute to cognitive impairment, termed "possible AD" in the NINCDS/ADRDA system, were also included. A total of 53 persons met dementia criteria at baseline and they were excluded from the analyses.

A set of 21 tests was administered as part of each evaluation. One test, the Mini-Mental State Examination,²⁶ was used only for descriptive purposes, and another test, Complex Ideational Material,²⁷ was not used in analyses because of a very skewed distribution. The remaining 19 tests were grouped into 5 domains of cognitive function, based in part on a previous factor analysis.²⁴ (1) Episodic memory: Word List Memory, Recall, and Recognition from the Consortium to Establish a Registry for Alzheimer Disease neuropsychological battery,²⁸ immediate and delayed recall of the East Boston Story,²⁹ and Story A from Logical Memory.³⁰ (2) Semantic memory: 20-item version of Boston Naming Test,³¹ Verbal Fluency,²⁸ 15-item version of Extended Range Vocabulary,³² 20-item version of National Adult Reading Test,³³ and subsequent modifications.^34- 35 (3) Working memory: Digit Span Forward and Digit Span Backward,³⁰ Digit Ordering,³⁶ and Alpha Span.³⁷ (4) Perceptual speed: Symbol Digit Modalities Test–Oral Version³⁸ and Number Comparison.³² (5) Visuospatial ability: 15-item version of Judgment of Line Orientation³⁹ and a 17-item version of Standard Progressive Matrices.⁴⁰

A composite measure of each cognitive domain was formed, as previously described,²⁴ by converting raw scores on the tests grouped in that domain to z scores, using the baseline mean and SD, and computing the average. To assess the differential association of ϵ4 with episodic memory compared with other cognitive domains, we computed the mean score in domains other than episodic memory to yield a nonmemory composite. Composite measuresto contrast with each of the other 4 cognitive domain measures were constructed in the same way, by averaging scores fromthe other domain measures.

Blood was collected at each participating Religious Orders Study site with acid citrate dextrose anticoagulant and stored at room temperature; it underwent lymphocyte separation within 24 hours of collection. DNA was extracted from approximately 2 to 3 million cells using a Puregene DNA isolation kit (Gentra, Minneapolis, Minn), with APOE genotypes determined according to the method described by Hixson and Vernier.⁴¹ Genotyping was done by an investigator blinded to all clinical data.

For all analyses, participants were divided into those with 1 or more ϵ4 alleles (ie, ϵ2/4, ϵ3/4, and ϵ4/4) and those without an ϵ4 allele (ie, ϵ2/2, ϵ2/3, and ϵ3/3).

We used a Cox proportional hazards model to assess how the presence of ϵ4 affected the risk of developing AD during the 6 years of observation, adjusting for the potentially confounding effects of age, sex, and education.⁴²

Random effects models were used to characterize individual paths of change in each cognitive measure and to examine how the ϵ4 allele was associated with initial level of function and annual rate of change.^24,43- 44 In this growth curve approach, each individual's path is assumed to follow the mean path of the group except for random effects that cause the initial level of function (ie, intercept) to be higher or lower and the rate of change (ie, slope) to be faster or slower. These random effects were assumed to follow a bivariate normal distribution. They were used to estimate individual growth curves, which were plotted. Model assumptions about linearity, normality, and independence and homoscedasticity of errors were assessed graphically and analytically, and found to be adequately met.

To assess the association of the ϵ4 allele with change in each cognitive domain, we constructed separate random effects models for each of the 5 cognitive domain measures. Each model included terms for time (in years since baseline), the presence of ϵ4, and their interaction. The term for time indicates the average annual rate of cognitive change in those without ϵ4. The term for ϵ4 indicates the average effect of ϵ4 on cognitive score at baseline. The interaction term indicates the average effect of ϵ4 on rate of cognitive change per year. Each model was also adjusted for the effects of age, sex, and education.

To test the hypothesis that the ϵ4 allele is selectively associated with loss of episodic memory, we conducted a series of analyses contrasting the episodic memory measure with a composite measure of the other cognitive domains. First, we constructed separate random effects models for the episodic memory measure and the nonmemory composite, each with a term for time, and estimated individual intercepts (ie, baseline level of function) and slopes (ie, annual rate of change) on each measure. Second, we constructed a model that used both sets of intercepts as outcomes. The model terms included an indicator of which outcome was being analyzed (coded 0 for the episodic memory intercept and 1 for the nonmemory composite intercept), ϵ4, and their interaction. Terms were also included to control for the effects of age, sex, and education. The interaction of the indicator with ϵ4 denotes whether ϵ4 was more strongly associated with baseline impairment in episodic memory than with baseline impairment in the nonmemory composite measure. Third, we constructed a similar model that used both sets of slopes as outcomes. In this model, the interaction of the indicator with ϵ4 denotes whether ϵ4 was more strongly associated with decline in episodic memory than with decline in the nonmemory composite.

To determine if cognitive systems other than episodic memory were selectively affected by ϵ4, we repeated the 3 analytic steps outlined for each of the remaining cognitive domain measures. In each case, the domain measure was contrasted with a composite measure of the other 4 domains.

The distribution of APOE genotypes in study participants was as follows: ϵ2/2 = 1; ϵ2/3 = 72; ϵ2/4 = 13; ϵ3/3 = 377; ϵ3/4 = 139; and ϵ4/4 = 9. This distribution is comparable with those reported in several population-based studies.^45- 48 Subgroups of those with or without at least 1 ϵ4 allele were similar in demographics and baseline Mini-Mental State Examination scores (Table 1).

During the 6 years of observation, 102 persons developed AD (92 with probable and 10 with possible AD). Their APOE genotypes were as follows: ϵ2/3 = 8; ϵ2/4 = 7; ϵ3/3 = 58; ϵ3/4 = 26; and ϵ4/4 = 3. Four persons developed dementia due to other causes (eg, stroke, Parkinson disease). A proportional hazards model was used to assess whether the expected association of ϵ4 with risk of developing AD was present in this cohort, excluding these 4 persons with other causes of dementia and adjusting for age, sex, and education. Compared with those without ϵ4, the relative risk of developing AD in those with ϵ4 was 1.92 (95% confidence interval [CI], 1.27-2.91). Comparable results were obtained when those with possible AD were excluded.

Table 2 presents the baseline mean and SD of each of the cognitive domain measures. We constructed separate random effects models to see how the presence of APOE ϵ4 was related to baseline level of function and annual rate of change in each domain after adjustment for demographic variables (Table 3).

In those without an ϵ4 allele, the episodic memory score declined an average of 0.021 standard units per year (95% CI, −0.026 to –0.016), as shown by the term for time. Those with ϵ4 had lower initial memory scores than those without ϵ4, by an average of 0.140 units (P<.01). In addition, those with ϵ4 declined more rapidly, by an average of 0.067 units (P<.001), which represents more than a 3-fold increase compared with those without ϵ4.

Those without an ϵ4 allele also declined in each of the other cognitive domains, with the average annual rates of decline ranging from 0.021 units (visuospatial ability) to 0.068 units (perceptual speed). Those with ϵ4 had lower visuospatial ability at baseline than those without ϵ4, by an average of 0.130 units (P<.05), but ϵ4 was not significantly related to baseline levels of semantic memory, working memory, or perceptual speed. Annual rate of decline was more rapid in all domains for those with ϵ4 compared with those without it, with increases of approximately 50% in working memory and perceptual speed and of approximately 150% in semantic memory and visuospatial ability.

Figure 1 shows the average paths of change in episodic memory (A) and a composite measure of the other 4 cognitive domains (B) during the 6 years of observation in the APOE subgroups. The deleterious effect of ϵ4 on episodic memory seems to be stronger and to emerge earlier than its effect on the nonmemory composite.

To test the hypothesis that APOE ϵ4 selectively affected episodic memory at baseline, we constructed a model that used baseline scores on both episodic memory and the nonmemory composite as outcomes. The model terms included an indicator of which outcome was being analyzed (ie, episodic memory or nonmemory composite), ϵ4, and their interaction. The estimate of this interaction term is presented in the first row of Table 4. It indicates that at baseline, the ϵ4 association with episodic memory impairment was marginally stronger than its association with impairment in the nonmemory composite (P = .06).

To test the hypothesis that ϵ4 selectively affected decline in episodic memory, we repeated the previous analysis, except that annual rates of change in episodic memory and in the nonmemory composite were used as outcomes instead of the baseline scores on those measures. The interaction of ϵ4 with the indicator term, presented in the second row of Table 4, was significant, denoting that ϵ4 had a stronger association with decline in episodic memory than with decline in the nonmemory composite.

To ascertain whether cognitive domains other than episodic memory were differentially affected by ϵ4, we repeated the same analytic steps for each of the other cognitive domains. Table 4 presents the key interaction term from each of the analyses comparing ϵ4 effects on a cognitive domain measure with its effects on a composite of the other 4 domains, and Figure 2 shows the average decline in each domain compared with the decline in the composite of the other 4 domains in the APOE subgroups. In semantic memory, ϵ4 had a weaker association with baseline level of the domain measure compared with the composite measure but ϵ4 was not differentially associated with annual change. Baseline function in the other 3 domains was not differentially associated with ϵ4 but rate of change was. In each case, ϵ4 had a weaker association with change in the domain measure than it did with change in the composite measure.

Because the APOE ϵ4 allele is associated with increased risk of AD, it must, at some level, also be related to decline in multiple cognitive systems. However, the extent to which ϵ4 has a relatively selective effect on episodic memory has been difficult to establish in previous research. In this 6-year longitudinal study of older persons, we found that possession of 1 or more ϵ4 alleles was associated with rate of decline in all forms of cognitive function. However, the effect of ϵ4 on change in episodic memory was much more pronounced. Further, at baseline, ϵ4 had a marginally stronger association with impairment in episodic memory compared with other cognitive functions, suggesting that ϵ4 effects on episodic memory precede those on other cognitive systems. Overall, these results support the hypothesis that ϵ4 has a relatively selective effect on episodic memory.

The biological mechanisms through which the ϵ4 allele affects cognitive function and the development of AD are not well understood. Some studies have reported an association between ϵ4 and level of AD pathology, particularly β-amyloid accumulation.⁴⁹ By contrast, APOE is also involved in neuronal repair and survival^50- 51 and in atherosclerosis,^52- 53 suggesting that ϵ4 may be related to disease through other mechanisms. Further, ϵ4 has been related to dementia in other diseases that are not typically associated with selective impairment of episodic memory. Although clinicopathological studies will be needed to more definitively address this issue, our results are consistent with the idea that ϵ4 influences risk of AD primarily by affecting the usual biological process that leads to AD rather than by some other mechanism.

Previous longitudinal studies have found that ϵ4 is associated with decline on tests of episodic memory,^4,15,17- 19 perceptual speed,^12,14,20 and global cognition.^{11- 12,16,18,20} Our results build on these findings by showing that ϵ4 is associated with decline in all cognitive systems but that its effect on episodic memory is relatively stronger than its effects on other forms of thinking.

Confidence in these findings is strengthened by several factors. First, there was an average of more than 5 evenly spaced observations per person, making it possible to reliably characterize change in individuals. Second, follow-up participation in survivors exceeded 95%, making it unlikely that attrition could substantially bias estimates of change. Third, composites of 2 or more individual cognitive function tests were used as outcomes, reducing the opportunity for floor and ceiling effects and other sources of measurement error to affect estimates of change. Fourth, the hypothesis that ϵ4 has a stronger association with episodic memory than with other cognitive functions was supported by both cross-sectional and longitudinal data.

Several limitations should be noted. First, findings in this selected group may not generalize to other groups. Participants were predominantly white, and there is evidence that APOE effects differ in other ethnic/racial groups.⁵⁴ On average, participants also differ from older people in the US population in education and lifestyle. It is possible that within-group similarities in lifestyle may have helped to highlight the behavioral effects of genetic differences by reducing the confounding effects of other environmental variables associated with cognitive decline and AD. Second, we cannot rule out the possibility that psychometric differences between the outcome measures may have contributed to their differential association with APOE ϵ4.

Accepted for publication January 7, 2002.

Author contributions: Study concept and design (Drs Wilson, Evans, and Bennett); acquisition of data (Drs Wilson, Schneider, Aggarwal, Cochran, Berry-Kravis, Fox, Evans, Bennett, and Ms Bach); analysis and interpretation of data (Drs Wilson, Barnes, Beckett, and Bennett); drafting of the manuscript (Drs Wilson, Aggarwal, Bennett, and Ms Bach); critical revision of the manuscript for important intellectual content (Drs Wilson, Schneider, Barnes, Beckett, Cochran, Berry-Kravis, Fox, Evans, and Bennett); statistical expertise (Drs Beckett and Bennett); obtained funding (Drs Schneider, Evans, and Bennett); administrative, technical, and material support (Drs Wilson, Barnes, Cochran, Berry-Kravis, Fox, Evans, Bennett, and Ms Bach); study supervision (Drs Aggarwal and Bennett and Ms Bach).

This research was supported by grants R01 AG15819, K08 AG00849, and P30 AG10161 from the National Institute on Aging, Bethesda, Md.

We are indebted to the altruism and support of the hundreds of nuns, priests, and brothers from the following groups participating in the Religious Orders Study: Archdiocesan priests of Chicago, Ill, Dubuque, Iowa, and Milwaukee, Wis; Benedictine Monks, Lisle, Ill, and Collegeville, Minn; Benedictine Sisters of Erie, Erie, Pa; Benedictine Sisters of the Sacred Heart, Lisle, Ill; Capuchins, Appleton, Wis; Christian Brothers, Chicago, Ill, and Memphis, Tenn; Diocesan priests of Gary, Ind; Dominicans, River Forest, Ill; Felician Sisters, Chicago, Ill; Franciscan Handmaids of Mary, New York, NY; Franciscans, Chicago, Ill; Holy Spirit Missionary Sisters, Techny, Ill; Maryknolls, Los Altos, Calif, and Maryknoll, NY; Norbertines, DePere, Wis; Oblate Sisters of Providence, Baltimore, Md; Passionists, Chicago, Ill; Presentation Sisters, BVM, Dubuque, Iowa; Servites, Chicago, Ill; Sinsinawa Dominican Sisters, Chicago, Ill, and Sinsinawa, Wis; Sisters of Charity, BVM, Chicago, Ill, and Dubuque, Iowa; Sisters of the Holy Family, New Orleans, La; Sisters of the Holy Family of Nazareth, DesPlaines, Ill; Sisters of Mercy of the Americas, Chicago, Ill, Aurora, Ill, and Erie, Pa; Sisters of St Benedict, St Cloud and St Joseph, Minn; Sisters of St Casimir, Chicago, Ill; Sisters of St Francis of Mary Immaculate, Joliet, Ill; Sisters of St Joseph of LaGrange, LaGrange Park, Ill; Society of Divine Word, Techny, Ill; Trappists, Gethsemani, Ky, and Peosta, Iowa; Wheaton Franciscan Sisters, Wheaton, Ill.

We also thank Todd Beck, MS, for statistical programming, and Carolyn DeVivo and Eithne Barton for preparing the manuscript.

Corresponding author and reprints: Robert S. Wilson, PhD, Rush Alzheimer's Disease Center, 1645 W Jackson Blvd, Suite 675, Chicago, IL 60612 (e-mail: rwilson@rush.edu).