Most studies on the possible role of environmental factors in Alzheimer's disease (AD) have used case—control methodology when seeking to relate differences in risk of AD to different exposures to putative environmental insults. Likely candidates include head injury (Mortimer et al, Reference Mortimer, French and Hutton1985, Reference Mortimer, van Duijn and Chandra1991); low social class, especially manual occupations (Reference Fratiglioni, Ahlbohm and ViitanenFratiglioni et al, 1993); and work with industrial solvents or printing (Reference Graves, van Duijn and ChandraGraves et al, 1991; Chen et al, Reference Chen, Dick and Seaton1999a ,Reference Chen, Wei and Seaton b ). With the exception of pre-dementia head injury (Reference Mortimer, French and HuttonMortimer et al, 1985) no single environmental factor has proved robust enough to be detected in several unrelated studies (Reference WhalleyWhalley, 1991). The proportion of cases in an early-onset AD sample that can be attributed to head injury is significant but not high, being estimated at around 4% of all cases (Reference Mortimer, van Duijn and ChandraMortimer et al, 1991).

Measures of social class in older people are difficult to make and interpret, especially in semi-skilled and manual workers, because of age-related social drift downwards, which is particularly evident in women. Occupational studies in AD are made more difficult because of the likely long lag period between the environmental exposures of interest and the late onset of dementia. In many countries, dementing illnesses do not present to clinical services and epidemiological studies must rely on specially recruited community-based, clinical or ‘at risk’ samples. Such recruitment is expensive and demanding of time, money and expertise. Not surprisingly, therefore, progress in understanding the possible role of the environment in AD remains slow.

More-rapid progress is being made in studies on the molecular basis and classification of AD. Although early-onset AD can occur as an autosomal dominant disorder, most instances appear to be sporadic with little evidence of genetic susceptibility. Nevertheless, Selkoe (Reference Selkoe1999) asserted that:

The evidence for any of these estimates is scanty. Most molecular genetic studies (quite rightly) focus on individuals selected from families multiply affected by AD. These studies have identified three genes in which mutations can directly cause AD. First, mutations in the amyloid precursor protein gene (APP) were reported (Reference Goate, Chartier-Harlin and MullanGoate et al, 1991; Reference Murrell, Farlow and ChettiMurrell et al, 1991; Reference Mullan, Crawford and AxelmanMullan et al, 1992), but these account for fewer than 0.1% of all early-onset cases. Two more implicated genes were identified on chromosomes 14 and 1 (Reference Rogaev, Sherrington and RogaevRogaev et al, 1995; Reference Sherrington, Rogaev and LiangSherrington et al, 1995), termed presenilin 1 and 2, respectively (PS1 and PS2). Christine van Broeckhoven and her colleagues (Reference Cruts, van Duijn and BackhovensCruts et al, 1998), surveying the frequency of mutant APP, PS1 and PS2 genes in a consecutive series of 101 early-onset AD cases, detected no APP mutations and only seven with PS1 or PS2 mutations. These provide the best estimates yet of the proportions of early-onset cases attributable to established mutations.

A fourth gene on chromosome 19 is also implicated in AD but not as a causal factor. This is the apolipoprotein E gene (APOE) locus at which a polymorphism (APOE ϵ2, ϵ3 or ϵ 4) is linked to expression of AD. The identification of the ϵ 4 allele as a risk factor for AD was established in genetic linkage and linkage disequilibrium studies (Reference Pericak-Vance, Bebout and GaskellPericak-Vance et al, 1991; Reference Saunders, Strittmatter and SchmechelSaunders et al, 1993; Reference Strittmatter, Saunders and SchmechelStrittmatter et al, 1993). The susceptibility to AD attributed to APOE ϵ4 is estimated at 50-60% (Reference Nalbantoglu, Gilfix and BertrandNalbantoglu et al, 1994) and remains the most robust finding in general population studies of AD. Age at onset, for example, is linked to AD in an ϵ 4 dose-dependent manner (Reference Corder, Saunders and StrittmatterCorder et al, 1993) so that the presence of two copies of the ϵ4 allele (ϵ4 homozygous) increases the risk of AD at an earlier age than the presence of a single copy (ϵ4 heterozygous). There is also some evidence of a protective effect of the ϵ2 allele (Reference Corder, Saunders and RischCorder et al, 1994). The strength of the evidence in support of a major role for APOE genotype is such that models of disease transmission within large samples of families have become both feasible and potentially informative.

Many commentators recognise that the risk of AD is determined partly by APOE genotype and partly by some other genetic and/or environmental factors (e.g. Reference Rao, Cupples and van DuijnRao et al, 1996). The effects of APOE genotype and family history of dementia on risk of AD were examined by Jarvik et al (Reference Jarvik, Larson and Goddard1996) in a complex segregation analysis of AD in 204 families. Within the limitations of what may have been too small a sample to be informative, the authors rejected all simple models of AD transmission. They concluded that more complex models were required to account for their observations and did not discount the contribution of multiple genetic factors or environmentally linked disease mechanisms. Models of disease transmission were also developed by Lindsay Farrer's group at Boston University School of Medicine to examine the contribution of APOE genotype and to account for the effects of gender. In what became the largest and most informative collaborative study of its kind in AD, Farrer and his colleagues (Reference Farrer, Cupples and van DuijnFarrer et al, 1995) assembled diagnostic and genealogical data on 549 AD patients and their first-degree relatives. The relationship between APOE status and gender was not straightforward. Among male relatives, the risk of AD in the ϵ 3/ϵ4 group was similar to that for the ϵ 3/ϵ3 group. Among female relatives, on the other hand, the risk for the ϵ3/ϵ4 group was almost twice that of the ϵ3/ϵ3 group. This difference suggests that gender may influence the risk of AD in ϵ4 carriers.

The same group subsequently demonstrated that in the presence of the ϵ 4 allele, a dominant model best fitted the data (Reference Rao, Cupples and van DuijnRao et al, 1996). Single-gene inheritance also best fitted the pattern of AD incidence in families lacking ϵ4, but the authors could not exclude more complex modes of transmission in these families. Their research on the influence of gender indicated that susceptibility to AD differed between men and women irrespective of APOE status. They concluded that AD is completely penetrant in women, but only 62-65% of men with a predisposing genotype developed AD. In part, the difference could be accounted for by the premature death of men who were ϵ4 carriers, as ϵ4 conveys susceptibility to cardiovascular disease (CVD) as well as to AD. The problem was addressed in a separate survival analysis of AD relatives, which showed that women had a higher risk of AD after adjustment for the effects of CVD deaths in men. This collaborative study presents what is probably the best example of a complex segregation analysis of carefully diagnosed AD patients. Epidemiological studies on late-onset AD meeting precise requirements were entered into a meta-analysis by Gao et al (Reference Gao, Hendrie and Hall1998) to determine the contributions of age and gender to AD. As reported for early-onset AD (Reference McGonigal, Thomas and McQuadeMcGonigal et al, 1993), Gao et al (Reference Gao, Hendrie and Hall1998) found an increased risk of dementia in women. The work of the Boston group provides some clues to the likely biological explanations of this excess.

GEOGRAPHICAL DISTRIBUTION OF EARLYONSET AD IN SCOTLAND

Differences in the geographical distribution of early- or late-onset AD are reported but are not well understood. The first largescale study identified differences between urban and rural populations, interpreted by Prince et al (Reference Prince, Cullen and Mann1994) as “the protective effect of a rural environment”. Earlier, Gurland et al (Reference Gurland, Cross and DeFiguerdo1979) had found some evidence for a higher prevalence of age-related dementia in New York than in London. Geographical differences are perhaps most striking in studies of indigenous Black Africans from tropical Africa. Neuropathological studies found no AD in western African samples of demented and non-demented old people (Reference Osuntokun, Ogunnyiy and LekwauwaOsuntokun et al, 1991).

Geographical differences between countries — between urban and rural areas and within a single large city — pose many questions in AD about exposure to putative harmful environmental factors, the frequency of mutant genes in ‘at risk’ populations, the role of susceptibility genes (such as APOE) and the possibility that some environmental factors (e.g. the rural diet) are protective. Farrer's group (Reference Farrer, Cupples and HainesFarrer et al, 1997) was the first to address the very real possibility that some exposures may censor ‘at risk’ populations, producing spurious gender effects. Their finding that the increased contribution of ϵ4 in women was not attributable to artefact is of considerable relevance to future studies in AD.

Scotland is well placed for the study of such complex interactions. Most of the population lives in a central belt running roughly west to east along a valley that divides the Southern Uplands from the Scottish Highlands. The western end of the belt contains over half the Scottish population in a conurbation of about 2.5 million people centred on the city of Glasgow. At the eastern end of the valley lies Edinburgh, the capital city. Traditional industries such as coal mining, steel production and shipbuilding have provided labour-intensive employment for most of the 20th century in the west. In the east, service industries and light manufacturing have predominated. Across this economic gradient in Scotland there is a well-established ‘health divide’. Age at death is lower, and death from vascular disease (notably cerebrovascular disease and myocardial infarction) more frequent in the west than in the east. These differences have been convincingly linked to socio-economic deprivation (Reference McLoone and BoddyMcLoone & Boddy, 1994).

During the period 1955-1974, almost all early-onset AD cases presenting in Edinburgh came under the care of Dr Elizabeth Robertson. She had made AD her special interest, and having established post-mortem facilities in the principal mental hospital, she obtained neuropathological confirmation in 95% of her cases. Non-random geographical distribution of early-onset AD was demonstrated within Edinburgh with the highest incidence in an inner-city tenemented area (Reference Whalley and HollowayWhalley & Holloway, 1985). Two apparently unrelated cases presented from the same apartment some 15 years apart. Previously, the pedigrees of all 69 cases had been examined in Register House, Edinburgh (Reference Whalley, Carothers and CollyerWhalley et al, 1982). None of the cases shared common ancestors and no family members, as far as could be ascertained from death certificates, had died of causes or in circumstances suggestive of dementia. Analysis of the public water supply in 1986 to each address of presentation detected no differences between areas (unpublished observations).

Having detected a non-random distribution of pathologically confirmed earlyonset AD in Edinburgh (Reference Whalley and HollowayWhalley & Holloway, 1985), we wished to examine this on a much larger scale. Specifically, we wished to test the proposed association between early-onset AD and aluminium content of the public water supply (Reference Martyn, Barker and OsmondMartyn et al, 1989), but this proved untestable largely because Scottish records revealed that the quality of the public water supply at any location has varied substantially over time. Variations arise partly because principal urban areas can switch between several alternative supplies as necessary, and partly because of continuous improvements in water quality and analytical techniques during the lifetime of our subjects. Our approach was to examine geographical variation in early-onset AD using small-area geographical analysis (Reference Whalley, Thomas and McGonicalWhalley et al, 1995a ). This relies on the division of the entire Scottish population into small areas (by postcode), each with about 5000 individuals. There was non-random geographical distribution of early-onset AD but not of the comparison conditions (vascular dementia, motor neuron disease, prostatic or ovarian cancers). These geographical differences could not be attributed to methodological artefact as other techniques of case-finding demonstrated the same differences between regions of Scotland. In the same study, there were geographical variations in early-onset vascular dementia and the frequency of cerebrovascular events, suggesting that some risk factors are shared between the two conditions (Reference Starr, Thomas and WhalleyStarr et al, 1996).

Non-random geographical variation in the incidence of early-onset AD in Scotland was mostly located in small areas of the country's densely populated central belt. These were concentrated in the large conurbations of Glasgow, the adjacent decaying industrial area of North Lanarkshire and relatively impoverished areas of Edinburgh adjacent to the Lothian coalfields. Rural parts of Scotland were relatively spared (Reference Thomas, Whalley and StarrThomas et al, 1997b ). Earlier reports (Reference Copeland, Gurland and DeweyCopeland et al, 1987; Reference Prince, Cullen and MannPrince et al, 1994) suggested that factors associated with urban life may increase the risk of early-onset AD, possibly through the same mechanisms that link urbanisation with increased incidence of cerebrovascular events (Reference Starr, Thomas and WhalleyStarr et al, 1996).

In a second analysis of the epidemiology of early-onset AD in Scotland, we related geographical variation to exposures to possible risk factors. Once more, we found that Scotland had been subject to detailed epidemiological analysis so that a great deal was known about the ecology of socio-economic deprivation in this country. Carstairs & Morris (Reference Carstairs and Morris1991) classified all of Scotland's small areas using a deprivation index based on an unweighted combination of four standardised variables (overcrowding, male unemployment, low social class and no car ownership). Socioeconomic deprivation measured in this way was significantly related to the incidence of vascular dementia but not to that of early-onset AD. Contrary to expectations concerning hazardous exposures in some industries, specific occupations were not overrepresented among men with early-onset AD.

Surprisingly, increased paternal age was associated with early-onset AD in men but not in women. An earlier small study of AD in Scotland (Reference Whalley, Carothers and CollyerWhalley et al, 1982) had detected increased parental age in AD. This observation was confirmed in other studies, but much later a larger study (Reference Farrer, Cupples and ConnorFarrer et al, 1991) reported decreased paternal age in late-onset AD. This latter study contained sufficient statistical power to distinguish between maternal and paternal contributions to parental age effects in dementia. In the Scottish studies, the most plausible biological explanation of the putative association between increased paternal age and the incidence of early-onset AD would be that fresh gene mutation is more likely to occur with advanced paternal age, as reported in some studies that distinguish between ‘sporadic’ and ‘familial’ instances of autosomal dominant disorders such as Marfan's syndrome (Reference EmeryEmery, 1986). Increased paternal age and paternal coalmining occupation acted independently of one another and were not associated with specific geographical areas at the time of presentation.

Epidemiological studies of this type which can draw upon all relevant health data and public record statistics for a defined period of time (in this instance 1974-1988) prompt exploration of the temporal associations between exposures to various risk factors. Elegant statistical models are now available to examine the putative relationships between birth characteristics and later environmental established risk factors for hypertension and vascular disease (Reference Elford, Phillips and ThomsonElford et al, 1990). When differences are detected between geographical areas and there is sizeable migration between those areas, this can provide useful insights into the likely timing and mechanism of the environmental effect (Reference BenthamBentham, 1988).

The Scottish studies therefore allow related hypotheses to be tested by examination of the time from clinical presentation to death (survival time) in small areas of high incidence compared with those of low incidence (Reference McGonigal, McQuade and ThomasMcGonigal et al, 1992; Reference Thomas, McGonigal and McQuadeThomas et al, 1997a ). In addition, public record data provided information about place and circumstances at birth and presentation to hospital services. The length of survival after presentation with early-onset AD distinguished between areas of high and low incidence, in that patients presenting in high-incidence areas had significantly shorter survival times than those presenting in low-incidence areas. Those born in high-incidence areas who presented in low-incidence areas survived as long as those who were born and presented in low-incidence areas. Those born in low-incidence areas who presented in high-incidence areas had a shorter survival time.

The worse prognosis for early-onset AD could be explained by one or more of several factors. For example, it could be associated with a more severe variant of the disorder (Reference Henderson, Jorm and KortenHenderson et al, 1992). If our measure of migration (the difference between place of birth and place of presentation) is valid and approximates to environmental exposures in adult life, then it becomes plausible to argue that the decreased survival rate in those who are either born in or move to areas of high incidence is explained by exposure to the harmful effects of the environment in the high-incidence area. This inference is supported by the observation that those born in high-incidence areas but who moved to low-incidence areas where they presented had much better survival, confidently linked to mitigation of the disease process. A similar finding was reported by Horner & Chirikos (Reference Horner and Chirikos1987) in their study of survival in gastrointestinal cancers. Here, the complexity of the statistical model becomes evident; high-incidence areas for early-onset AD may also be characterised by increased incidence of other diseases that can cause premature death.

GENETIC AND ENVIRONMENTAL RISK FACTORS: POSSIBLE LINKS

Small areas of high incidence of early-onset AD may be explained by genetic relatedness such that mutated causal genes or genes that convey increased susceptibility to this disorder are more frequent in these localities. However, geographical variation related to locally increased gene frequency should be most obvious at time of birth and less obvious with the effects of migration over time. One of the limitations of our statistical analyses was that data on place of birth were insufficient to allow an informative analysis of geographical variation. However, other factors present at time of birth (paternal age and occupation) certainly contributed to the increased risk of early-onset AD. When we estimated the aetiological fraction attributable to other established risk factors for AD we calculated that the proportion attributable to being female was 25%, to high paternal age 9.3% (male patients only) and to having a coal-miner father 10% (both genders). These factors together appear to act independently and contribute 45% to the total. A further 20%, we estimated, was attributable to environmental exposures in later life.

CONCLUSION

Brain diseases, age-related brain degeneration and neurodevelopment are discrete independent processes that are locked together in complex ways. There are varied influences on the processes of neural differentiation, maintenance and repair in the face of age-related disease and/or degeneration. A comprehensive multi-factorial model of early-onset AD must distinguish between genetic factors related to ancestral genes, fresh gene mutation, gender and agerelated genes, the harmful effects on neurodevelopment of privations such as malnutrition in early life, and exposure to a harmful environment in mid-life. If the dietary environment includes nutrients with known advantages for nervous development (e.g. essential fatty acids in the early diet, or antioxidants in fresh fruit and vegetables in late diet), then the onset of a dementing illness may be postponed. If it is also supposed that age at onset of AD is at least partly determined by exposure to a harmful environment in early life, this could implicate a reduction in the ‘cerebral reserve capacity’ available to adapt to insult, accelerating the onset of symptoms of dementia.

This brief review has focused on some of the most pressing problems in understanding the occurrence of AD. Genetic factors are of established importance and are intensively studied. The interaction between APOE status and gender provides a useful insight into the complexity of the issues and informs future researchers of the scale of effort required to disentangle the genetic and environmental factors involved. Despite their considerable effort in recruiting over 600 families, Lindsay Farrer's group expressed doubts that their sample size might still be too small to disentangle the complex factors involved (Reference Farrer, Cupples and van DuijnFarrer et al, 1995). Geographical distribution of a Scottish national sample of AD with post hoc kinship analysis of disease clusters provided a complementary approach to classic segregation analysis in large samples. Although each technique has its disadvantages, ecological data used alone will always be much less exact than those based on more direct study. However, some combination of the two approaches, with the added precision afforded by molecular genetic analysis, may provide the most informative solution. Such studies are now under way in at least three major centres worldwide. Their findings will help answer some of the many unanswered questions about the causes of AD. It is no longer speculative to propose that future clinical practice in dementia will be influenced by the results of epidemiological enquiry. Much as has occurred in the prevention of cardiovascular disease, individuals entering the risk period for dementia will be offered an assessment of their risk of developing the condition. The techniques of epidemiology will be in considerable demand to evaluate and refine the interventions thought likely to delay or even prevent the onset of dementia (Whalley & Struth, Reference Whalley and Struth1998, Reference Whalley and Struth1999; Reference SelkoeSelkoe, 1999).