In the only detailed population-based studies from sub-Saharan Africa, the prevalence and incidence of Alzheimer's Disease (AD) and dementia in Nigeria are very low [1, 2]. However, among African-Americans, prevalence and incidence rates are similar to [1, 2], or even higher [3, 4] than the rates for white non-Hispanic Americans. Also, the prevalence of dementia in some Caribbean and South American populations with African admixture, are among the highest in the world [5–8].

The e4 allele of the apolipoprotein-E gene was first reported to be associated with an increased risk of AD twenty years ago [9, 10]. Since then, this has been the most consistently replicated genetic risk factor [11]. The association has been observed in many different populations [12]. However, in African-Americans, other populations of west African ancestry, and Hispanics the association of AD with e4 is relatively weak and inconsistent, even though the frequency of the risk-conferring APOE e4 allele is higher in those of African ancestry than in other continental groups [13]. No association of APOE genotype with AD has been observed in studies in Nigeria [14], or Kenya [15]. Two longitudinal studies in the US also suggest no association between APOE e4 and incident AD among African Americans, while the incidence of AD seemed to be higher for African Americans in every APOE genotype [3, 4]. In a case-control study in Florida the association between APOE e4 and AD was as strong for Cuban Americans as for white non-Hispanics [16] in contrast to the absence of an observed association among Hispanics in North Manhattan [4]. This pattern of findings is strongly suggestive of the presence of gene by environment, and/or gene by gene interactions [17].

Those classified in the US as 'Hispanic' originate from diverse mixed ancestry Caribbean, Central and South American populations, resulting from two-way admixture between Native American and European populations or three-way admixture among Native American, European, and West African populations [18]. However, patterns of admixture vary greatly among these populations. The catch-all 'Hispanic' category is therefore problematic, providing some information about linguistic and cultural heritage but very little about ancestry. In much of continental Latin America, two-way admixture dominates with little evidence of African ancestry [19]. Cuba is quite different. The first European contact was by Columbus in 1492. The indigenous population was reputedly extinct by 1700. While the ancestral Native American substrate is still appreciable in the maternal lineages, the extensive process of population admixture in Cuba has left no trace of paternal Native American lineages, mirroring the strong sexual bias in the admixture processes taking place during colonial times [20]; currently Native American admixture is minimal [21]. Importation of slaves from West Africa was current by 1600 and not abolished until 1886. The proportion of population identified as black or mixed rose from 34% in 1774 to 57% by 1817. Recent Cuban studies concur in identifying average proportions of African admixture in those who classify themselves as white, mixed race and black as, respectively, about 5%, 35% and 60% [20, 21]. European admixture among African-Americans is much lower, an average of between 12% and 20% in different US cities [22] and very few African-Americans have as much as 50% European ancestry [23]. In the former British Caribbean, average European admixture levels may be even lower; just 7% in Jamaica [22].

The high levels of African and European admixture in Cuba can be used to good effect. Studying the relationship of dementia risk to individual admixture within admixed populations is the most direct way to distinguish genetic from environmental explanations for ethnic differences in disease risk [24], and, by extension, for distinguishing gene by environment versus gene by gene explanations for ethnic differences in the effects of genes on disease outcomes. Furthermore, such relationships will confound studies of other genetic risk factors - "hidden population stratification". Measurement of the confounder (individual admixture) allows us to control for population stratification using standard methods.

We recently reported a high prevalence of dementia in a one phase prevalence study in an older Cuban urban population [25]. In the present study we aimed to

1. analyse the association between ethnic group and individual admixture

2. assess the association of each with APOE genotype

There has been much interest in the potential role of African ancestry in modifying the effect of the APOE genotype and influencing risk for AD and other dementias. We believe this to be the first study to have addressed this issue directly, through estimation of individual admixture, rather than relying merely on observations of ethnic type. Another strength is the population-based survey design with high response rates for the main survey and blood sample collection. The main weaknesses of the study were, first, that the sample size, although large, may not have been adequate to exclude important APOE genotype by admixture interaction effects, and second that in the cross-sectional design we were unable to distinguish environmental factors associated with the incidence of dementia from those predicting its duration. Finally, although the use of 60 SNPs to estimate individual admixture is considered to provide reasonable precision, a standard error of around 0.1 is to be expected. Ideally we should have accounted for the uncertainty in these estimates within the regression analysis, rather than treating it as a covariate observed without error. We do not believe that this will have led to systematic error, since most SNPs were successfully genotyped on most participants. Nevertheless, although much more computationally demanding, future more definitive tests of these hypotheses on larger samples would benefit from this more rigorous approach

In this representative population-based survey of older Cubans in Havana and Matanzas cities, ''white', mixed' and 'black' ethnic groups were all substantially admixed, with varying proportions of African and European ancestry. There was a strong and statistically significant association between both ethnic identity and admixture and the APOE genotype, the e4 allele being over-represented in 'mixed' and 'black' ethnic groups and in those with greater African admixture. Overall, we found a strong association between APOE genotype and dementia, with effect sizes very similar to those reported in other settings [11, 13]. The association was evident among those identified by interviewers as 'black' as well as those identified as 'white', but not in those identified as 'mixed'. The 'mixed' group was the smallest in number, and lack of precision may have contributed to this otherwise surprising finding. There was a non-significant trend for the association between APOE genotype and dementia to be weaker in those with greater degrees of African admixture. Controlling for ethnic identity or admixture did not affect the association between APOE genotype and dementia, suggesting an absence of confounding by population stratification. After controlling for compositional differences in APOE genotype (the risk conferring e4 allele being more common in 'mixed' and 'black' ethnic identity groups), there was a non-significant trend towards lower dementia prevalence in those 'non-white' groups. A similar non-significant trend was apparent for admixture. However, when the joint independent effects of ethnic identity and admixture were assessed in a single model, mutual confounding was evident. In each ethnic identity, increased African ancestry greatly increased the risk of dementia. At every level of African ancestry, those with 'mixed' and 'black' ethnic identities had a lower risk of dementia

We have established a link between admixture and APOE genotype, with a higher frequency of the risk-conferring e4 allele in those with greater degrees of African admixture. All things being equal, this would be expected to result in a greater incidence and prevalence of dementia. However, in our sample this was offset by a large attenuation of the effect of APOE e4 in those with more African ancestry. This interaction was not statistically significant, and larger samples will be required to measure this with more precision and exclude type II error. Also, there was no significant graded effect modification by ethnic identity in the larger sample, with the attenuation of effect being confined to those in the 'mixed' group. Gene by environment interactions still seem the most plausible explanation for apparent modification of the effect of APOE among African, African American and European populations, given the large difference in environmental exposures, particularly cardiovascular risk factors, between African and African American populations [36]. However, European admixture among African American populations may have also created potential for differential gene by gene interactions between the two settings. Were a more consistent and unequivocal interaction with admixture to be demonstrated in larger samples, we could then in principle begin to localize the genes responsible by admixture mapping, exploiting information about linkage generated by admixture [35].

Another balancing effect on overall prevalence may be in operation given that the effects of 'mixed' or 'black' ethnic identity on the one hand, and African genetic admixture on the other seem to be operating in opposing directions in influencing dementia risk. These are related yet by no means collinear constructs; hence mutual confounding is feasible. The new respectability of observer assessments of 'ethno-race' in epidemiological research arise precisely from their ability to identify the externally observable physical characteristics that are hypothesised to lead to social, economic and health disadvantage. Their utility in research, as well as their limitations are neatly summarised by an American epidemiologist, Camara Phyllis Jones [37]:

'The race that we measure in our studies is the same race that is noted by a taxi driver, a police officer, a judge in a courtroom, or a teacher in a classroom. That is, race is a social classification in our race-conscious society that conditions most aspects of our daily-life experiences and results in profound differences in life chances. This assigned race varies among countries. For example in the United States I am clearly labelled Black, while in Brazil I would be just as clearly labelled White and in South Africa, I would be clearly labelled 'coloured'. It is likely, if I stayed long enough in one of these settings, my health profile would become that of the group to which I had been assigned, even though I would have the same genetic endowment in all three settings.'

Much research in the US has focussed upon black ethnic identity as a socially determined, contextually bound construct, linked to disadvantage and discrimination, and mediating health disadvantages. Thus, in the 1990s darker skin colour among African-Americans was found to be inversely associated with income, education and occupational status, and to be a stronger predictor of adult occupational status than was parental socioeconomic position [38]. More recently, darker skin colour has been shown to be independently associated with experiences of racism [39]. Protective income gradients in hypertension are evident for light-skinned but not dark-skinned African-Americans, an effect hypothesised to be explained by psychosocial stressors linked to skin colour, including racism [40]. Of relevance to our finding of a protective effect of non-white ethnic identity on dementia risk, some benefits of such self-identification have been reported; for example factors reflecting participation in and belonging to African-American culture were associated with a range of positive health behaviours [41].

Genetic admixture, externally observable physical characteristics including skin colour, and self-reported ethnicity are related to each other, but in complex ways [42, 43]. One of the weaknesses in the current study is that we did not adequately separate out self-perceptions from observer ratings of ethnic identity. A proper understanding of the role of genetic admixture in determining disease risk may require measurement of, and control for each of these and other related socio-cultural factors, including socioeconomic position and acculturation [37, 44]. Although strongly correlated, the effects of admixture and ethnic identity on health outcomes can and should be distinguished when assessing genetic and environmental contributions.