APOE Polymorphisms and Late-Onset Alzheimer Disease: Title and subTitle BreakThe Importance of Ethnicity

Ethnic groups traditionally have been underrepresented in research, and research in diseases of aging is no exception. To increase recruitment of underrepresented ethnic groups, the National Institute on Aging has provided supplementary funds to Alzheimer disease (AD) centers around the country for the purpose of recruiting cohorts of patients and controls. In this issue of THE JOURNAL, we see an example of the fruits of such research with these patient populations. Tang and colleagues,¹ who have helped pioneer the development of epidemiologic research on late-life dementias, report that African Americans and Hispanics living in Manhattan have a higher relative risk of possible or probable AD than whites. Unlike their white neighbors, however, there is little evidence that their risk of developing this devastating disorder is enhanced by carrying 1 of 3 relatively common variants of the gene that codes for apolipoprotein E (APOE). That variant, the ∊4 variant (APOE-∊4), was first shown to be a risk factor in 1993² and has since been confirmed in numerous other populations.³

While methodologic issues must always be addressed and discussed in the interpretation of any such study, the conclusion of Tang et al that the risk of AD varies by race and ethnicity, regardless of APOE genotype, has a number of important implications. We do not know, of course, if the reported differential incidences of AD and susceptibilities to APOE polymorphisms are functions of background genotype, environment (including early development), or both. Regardless of the relative roles of nature and nurture, however, the findings shed light on our understanding of how AD develops. The current dogma regarding pathogenesis invokes altered metabolism of the β-amyloid precursor protein as the key culprit.⁴

The literature on different isoforms of the APOE protein generally points to evidence of variable interactions with β-amyloid precursor protein, but this evidence is uneven.⁵ Moreover, the "bad" APOE (derived from the APOE-∊4 allele) can influence one's risk for a variety of other neurodegenerative disorders,⁶ while the "good" APOE (derived from the APOE-∊2 allele) increases the risk of cerebral hemorrhage associated with β-amyloid deposits within cerebral blood vessels.⁷ Perhaps APOE isoforms interact at variable levels of efficiency with the ciliary neurotrophic factor, an important neurotrophin that could determine how well the brain recovers from a variety of injuries.⁸ The situation is even more complex: polymorphic variants within the promoter of APOE, presumably conferring different regulatory properties, can influence risk of AD independently of the effect of APOE-∊4.⁹ How this new risk factor plays out among various ethnic populations will be interesting to discover.

Tang et al¹ accrued a total of 221 AD cases during follow-up of their cohort of 1079 persons (53 African Americans, 23 whites, and 145 Hispanics). They examined the strength of association between genotypes containing APOE-∊4 and AD and found relative risk estimates consistent with those reported in the literature, but only for whites. Hispanics and African Americans showed no significant increase in risk of AD associated with APOE-∊4. In contrast, among those with non-∊4 genotypes, Tang et al found a roughly 2-fold increase in AD risk for Hispanics and a 4-fold increase for African Americans compared with whites. These findings reportedly withstood adjustment for a variety of potential confounding factors, including age, sex, education, family history of dementia, and hypertension. While the confounding effect of education was adjusted for, no information was provided concerning possible differences in the quality of education. Quality could differ among ethnic groups and, because of the putative association between low levels of education and AD, could conceivably modulate susceptibility to AD. However, it is unlikely to account entirely for the magnitude of the observed association between ethnicity and AD, and it does not appear to affect the association between ethnicity and AD in the same direction among those who have the APOE-∊4 allele. Application of survival analysis techniques demonstrated that the increase in relative risk for African Americans and Hispanics (among those without an ∊4 allele) persisted to age 90 years.

The strength of this investigation lies in its focus on relatively understudied groups in a prospective study design. To date, few other studies have addressed the question of AD incidence and genotype in this way. Like most racial/ethnic groups, African Americans constitute a nonhomogeneous population in terms of genetic constitution, and it may be difficult to identify and study finer classifications of the group. Within ethnic groups, testing for Hardy-Weinberg equilibrium reduces but does not totally eliminate concerns that observed allele frequencies are unrepresentative, because of recent genetic admixture or other factors.¹⁰ Similarly, Hispanics were classified as such based on self-report, after they had already responded to the US Census question about their race.¹ Thus, subjects in the Hispanic group may have been included primarily because of their language or birthplace. Despite the limitations and imprecision of these ethnic classifications, such analyses can provide important insights into AD risk factors.

How subjects were enrolled and how they were retained in the study following enrollment can influence how we interpret these results. Of 4865 persons identified, 2128 were initially interviewed. In a cohort study such as this, the initial participation rate primarily affects study generalizability, while subject retention primarily affects the internal validity of the study. The initial participation rate can affect internal validity, but only when participation is differentially associated with both risk factors and outcome.¹¹ It seems unlikely that subjects in one ethnic group would decide to participate because of their (unknown) genotype and potential future risk of AD, so the initial participation rate would not be expected to be a major source of bias. Generalizability, on the other hand, concerns whether the relationship between APOE genotype and AD can be expected to be associated in the same way in Hispanics or African Americans who were not included in this particular study. Generalizability is a matter of judgment. For example, if an internally valid cohort study was conducted in men and demonstrated that smoking was associated with lung cancer, would it be appropriate to generalize that smoking might also cause lung cancer in women or in other men outside the study cohort? Because of uncertainty of the pathogenesis of AD as well as the potential heterogeneity of Hispanics and African Americans, generalizations concerning these data should be made cautiously.

In the usual cohort study design, persons free of the disease being studied are enrolled, and attempts are made to incorporate person-time contributed by those who die or are unavailable for follow-up as well as those who contract disease or remain disease free at the time of analysis. Bias could be generated if rates of being unavailable for follow-up or refusal to participate further after entry into the cohort differed by ethnic group, AD, or APOE genotype. Such bias could affect the overall conclusions. For example, suppose that death or being unavailable for follow-up did not occur evenly or randomly across ethnic groups, but instead was more frequent in one group than another. This could bias the calculated relative risk, threatening the validity of the results. In the study by Tang et al,¹ similar proportions of African Americans and whites refused genotyping or were unavailable for follow-up; Hispanics had fewer losses in these categories. Thus, bias caused by differences in rates of being unavailable for follow-up is unlikely to explain the observed differences in relative risk reported for Hispanics and African Americans (without the APOE-∊4 allele), unless the differential frequency of AD also occurred in those unavailable for follow-up. For bias to be the only explanation, whites who would later become demented would have to drop out, and/or African Americans who would later become demented would remain in the study.

The data presented by Tang et al¹ on numbers of subjects, mean duration of follow-up, and accrued cases appear to indicate a very high incidence of AD in African Americans. However, person-time contributed by censored observations is not explicitly included in the tables. Thus, incidence rates probably cannot be accurately estimated from the data presented. The censored observations, however, appear to be reflected in the cumulative incidence curves presented in the figures. These data suggest that, among persons without the APOE-∊4 allele, being African American may be one of the strongest risk factors for AD yet observed. The strength of this association is comparable to that attributed to the APOE-∊4 allele in other populations, but the point estimate of the relative risk lacks precision (95% confidence interval, 1.6-12.4) because of the relatively small number of subjects. Further independent replication of these findings is necessary before any definitive conclusions can be made regarding ethnic variation in the risk of AD. Nonetheless, the results reported by Tang et al should definitely spur further research in this relatively understudied area.

Demonstrating definitively an elevated risk for certain ethnic groups after eliminating the effect of the principal known genetic risk factor (APOE-∊4 allele) would have a substantial impact on the epidemiology of AD. Thus, the results presented by Tang et al¹ are at once provocative and exciting. Close examination of high-risk groups can increase the likelihood that new risk factors will be identified. What other potential risk factors should the scientific community now look toward to explain the potential difference in cumulative incidence among whites, African Americans, and Hispanics? One area for future investigation is the possibility that another gene (or genes) is associated with AD in these ethnic groups. Furthermore, the search for factors associated with the environment and with the early life history, as well as gene-environment interactions, may be enhanced by the study of populations with relatively high risks of AD. Finally, epidemiologic studies would also be enhanced by thorough neuropathologic analyses to further address the issue of heterogeneity.