Three Decades of Dementia Research: Insights from One Small Community of Indomitable Rotterdammers

DEMENTIA FROM A POPULATION PERSPECTIVE: PRESENT AND FUTURE

The inception of the Rotterdam Study cohort in 1990 provided the first population-based numbers of dementia prevalence in the Netherlands [1], contributing at the time to early efforts in Europe and the US to reliably map the prevalence of dementia in the population [2]. Over 15,000 people from Ommoord, a suburb of the city of Rotterdam, now participate in the ongoing study. Apart from the 4-yearly visits to the research center, participants are under continuous surveillance via the medical records of their general practitioner (a ‘gatekeeper’ in the Dutch healthcare system). This provides important information on their wellbeing, even when frailty prevents repeated assessment at the study center, allowing reliable estimates of lifetime risks and life expectancy with disease. Roughly one in three individuals in the Rotterdam Study get diagnosed with dementia, stroke, or parkinsonism during their lifetime, and roughly 1 in 5 develop dementia, with an average age at diagnosis of 81 years [3]. In terms of life expectancy, this means that on average 6% of remaining life years at age 65 will be lived with dementia, increasing to over 35% at age 95. Such findings are needed to guide formation of health care policy and prioritizing research efforts.

Much of the recent thrive in dementia research, and its funding [4], is sparked by ‘epidemic’ projections, with global prevalence expected to triple by 2050 [5]. But how reliable are the forecasts? Following initial prevalence estimates, the long-term, methodologically consistent observations of the Rotterdam Study cohort have enabled assessment of trends in de occurrence of dementia over the past 27 years. In support of evidence from Rochester (Minnesota) in the United States (US) [6], we showed in 2012 that the age-specific incidence of dementia has in fact been declining [7], accompanied by larger brain volumes and lower burden of white matter hyperintensities. Alike the first reports about the rise and fall of myocardial infarction incidence in the middle of the 20th century, this triggered much debate, which is still largely unresolved in terms of factors underlying these trends [8, 9]. Nevertheless, various studies have now corroborated the declining incidence trends, at least in the US and Europe [8], highlighting that developments over the past century, be it in educational attainment, treatment of cardiovascular risk factors and diseases, or other public health developments like hygiene, have benefitted our resilience against dementia. Yet, these optimistic trends do not negate the projected growth in the number of people living with dementia due to the ageing population, and may even be offset by increases in the prevalence of obesity [10], type 2 diabetes [11], and hypertension [12]. This illustrates the importance of identifying modifiable risk factors, and unravelling their role years, if not decades, before the onset of clinical symptoms of dementia.

MODIFIABLE RISK FACTORS: KEY TO CURBING THE EPIDEMIC?

Designed to identify determinants of disease and disability in the elderly [1, 13], much of the research done in the Rotterdam Study over the past years has been aimed at identifying modifiable risk factors for cognitive decline and dementia. Unlike biomarkers, which often reflect early subclinical alterations due to the disease process and are discussed below, the identification of causally related risk factors has the ultimate aim of developing preventive interventions. The Rotterdam Study, along with other population-based studies, has shown that known modifiable risk factors are accountable for approximately 25–30% of all dementia cases [14, 15]. By calculating the so-called population attributable risk, which takes into consideration the overlap between risk factors in individuals, this implies that eliminating these risk factors from the population would reduce the incidence of dementia by nearly a third. Although complete elimination of a risk factor from the population is often unachievable, it illustrates the large burden of these well-known risk factors on public health. Albeit generally modest in effect size at the individual level [16], public health interventions that target these risk factors could greatly reduce the burden of disease at the population level. The exponentially increasing incidence of dementia with age, unlike any other disease, has large implications for the potential of preventive medicine. The vast majority of life years spent with dementia are lived in the final few years of one’s lifespan, meaning that postponing the onset of dementia by merely a few years can reduce the lifetime risk and number of life years spent with dementia by up to 50% , as seen in the Rotterdam Study and beyond [17].

Despite the generally late-life onset, dementia is increasingly becoming ‘a disease of mid-life’, which reflects the general uncertainty regarding the earliest origin of the disease. Various risk factors, notably obesity [18] and hypertension [19], are particularly detrimental to late-life cognition when present in mid-life. The age at which people are eligible for the Rotterdam Study has dropped from≥55 in the inception cohort to≥40 years in the latest inclusion wave to reflect the importance of life course data [1]. Trajectories of cholesterol and blood pressure levels [20], but also clinical manifestations like depressive symptoms [21], aid in disentangling the time course and thereby mechanisms by which these are related to dementia onset. Moving onward by going further back in time, reliably tracking the life course of individuals from the prenatal phase and childhood [22] to adulthood and late-life is the next step in understanding of dementia, by linking developmental variation to neurodegenerative sequelae.

Given the importance of vascular disease in the onset of dementia, it is well conceivable that patients with heart disease are at increased risk of dementia. Thanks to improving acute treatments and secondary prevention, many patients with coronary heart disease or heart failure now live well into old age and are consequent susceptible to late-life diseases like dementia. Indeed, a history of heart disease relates to an increased risk of developing dementia, apparently independent of aforementioned risk factors, and even in the presence of subclinical myocardial infarction or cardiac dysfunction [23, 24]. This puts forward the possibility that long-term hemodynamic impairment and consequent hypoxia could be detrimental to brain health. We recently observed in the Rotterdam Study that low cerebral perfusion increases dementia risk, as well as cognitive decline in non-demented individuals, during on average 7 years of follow-up [25]. Although this does not rule out reduced perfusion due to metabolic changes in the very early preclinical phase of dementia, it does support further studies to assess whether improvements in perfusion, for example by physical activity, could be beneficial to brain health. In this context, cerebral autoregulatory mechanisms, including vasoreactivity and autonomic function [26, 27] could be vital to maintain sufficient oxygenation in the presence of disturbed flow. Further (circumstantial) evidence for a role of hypoxia in dementia etiology comes from associations of hemoglobin levels with cerebral perfusion, and long-term risk of dementia possibly pointing to a regulatory mechanism which involves either to maintain tissue oxygenation [28].

Other projects have focused on the role of lifestyle factors, beyond traditional cardiovascular risk factors, in the development of dementia. Educational attainment [14], depressive symptoms [14], and hearing loss [29], are examples of modifiable factors that may contribute to prevention of dementia [30]. In addition, physical activity is widely regarded as a protective factor against dementia. In part, this is supported by observations in the Rotterdam Study, showing protective associations up till 5 years of follow-up, but not thereafter [31]. This time-window raises the possibility of reverse causation, but could also be due to changes in behavior otherwise, the limitations of a single measurements, and the coarse nature of a physical activity questionnaire. In terms of diet, a healthy diet is generally considered beneficial, but lack of association between dietary adherence and cognitive decline [32], and dementia [33], which we observed in Rotterdam as well as in observational studies, suggests that there may be specific components only, such as represented in the Mediterranean diet [34], that have beneficial effects on cognition. In addition, observational studies vary widely in terms of intensity, frequency, and duration of exposure [35], highlighting the need for standardized quantification criteria for diet, as well as other of aforementioned lifestyle factors.

The challenge before us is to translate these observations on risk factors into biological mechanisms, in other words to treat risk factor associations as probes to advance etiological insight and ultimately facilitate preventive interventions. This regularly requires collaboration with basic and translational science, whereas advances in – omics initiatives (e.g., genomics and metabolomics) have created new opportunities to translate observations in the population to plausible mechanisms. Also, emerging measurement techniques, i.e., imaging, make it possible to examine community-dwelling individuals in more depth. The yield and further implications of these developments for population-based studies like the Rotterdam Study will be discussed in the next sections.

PRECLINICAL IMAGING IN DEMENTIA: WHAT HAVE WE LEARNED?

Already in the early 1990s, the Rotterdam Study acknowledged that thorough investigation of the etiology and the pathological mechanisms of dementia required in-depth visualization of the brain. The Rotterdam Study has always been a pioneer in introducing state-of-the-art imaging techniques into the population-based setting. These include retinal imaging, computed tomography imaging, magnetic resonance imaging, diffusion tensor imaging, and resting state imaging and the current section is dedicated to several of the most important recent findings from the Rotterdam Study [1, 36].

Brain imaging represents one of the cornerstones of current-day population-based dementia research [37]. As one the first population-based studies worldwide, the Rotterdam Study incorporated brain magnetic resonance imaging (MRI) into the core study protocol in 2005 [36]. Additionally, already in 1990 and then again in 1995 and 1999, subsets of individuals were invited for MRI. The excellent capacity of MRI to visualize brain structure and pathology, combined with its ability to assess brain function using the properties of the cerebral circulation, positions MRI as an extremely valuable non-invasive imaging tool for population-based brain imaging.

A strong focus of the Rotterdam Study has always been to establish cerebral small vessel disease as an important substrate of the dementia process. Among the earliest findings, we showed that a larger burden of white matter hyperintensities, lacunes, and cerebral microbleeds are associated with a higher risk of dementia and mortality [38, 39]. Importantly, these markers of cerebral small vessel disease are also directly linked to the preclinical phase of dementia, evidenced by the associations with cognitive deterioration, mild cognitive impairment, and deterioration of and impairment in daily functioning [39–41]. Of note, these pathologies to accumulate more in the white matter than in the grey matter. A natural extension was therefore to identify even earlier markers of presumed vascular damage. This led to the introduction in 2005 of diffusion tensor imaging (DTI) into the MRI protocol. Indeed, microstructural changes as quantified using DTI, were found to be already present in the normal appearing white matter and to precede the formation of white matter hyperintensities [42]. These findings further established what was already known through small clinical and animal studies: white matter hyperintensities develop gradually and that those that are visible only represent a small portion of the underlying white matter pathology. In terms of clinical relevance, we described that general loss of the microstructural integrity of the white matter plays an important role in the etiology of cognitive impairment and dementia [43, 44]. Interestingly, we even showed that degenerative changes in white matter microstructure also mark health outcomes beyond the brain, including gait impairments and a higher risk of all-cause and cardiovascular mortality [45, 46].

In addition to assessments of general microstructural degeneration diffusion-weighted imaging also allows quantification of microstructural integrity of specific white matter tracts [47]. Using these techniques, we found differential patterns of degeneration in specific white matter tracts with aging. In particular the limbic, association, and commissural tracts, appeared to degenerate most prominently with aging and might represent more specific neurodegenerative markers than overall white matter atrophy or the amount of white matter hyperintensities [44, 48].

Given our strong focus on the vascular pathways underlying dementia, the Rotterdam Study introduced various other modalities that comprehensively probe the cerebral microvasculature and hemodynamics. We discuss here cerebral perfusion, computed tomography (CT) imaging of the cerebral arteries, and retinal imaging.

With the introduction of our dedicated MRI-scanner in 2005, we also introduced measurements of total cerebral blood flow (i.e., cerebral perfusion) with use of dedicated phase contrast sequences [36]. In one of the earlier studies on cerebral perfusion a direct link between reduced perfusion and degenerative brain changes was shown [49], with complex underlying associations between cerebral blood flow and brain [50]. In terms of clinical significance, we also recently highlighted that cerebral hypoperfusion is associated with accelerated cognitive decline and an increased risk of dementia [25].

Toward the end of the millennium, conventional ultrasound of the carotids was the hallmark of large-vessel damage as it relates to dementia [51–53]. In 2002, the Rotterdam Study moved beyond ultrasound and added CT-imaging in a large subset of the population. In contrast to other population-based studies that introduced CT-imaging around the same time focusing on the coronaries and aorta, the Rotterdam Study had a broader scope and included visualization of intracranial vessels [54]. In several papers since, we showed that calcification in the extracranial and intracranial vessels contributes to cerebral atrophy, cerebral small vessel disease, cognitive decline and dementia [55, 56], further emphasizing the importance of vascular disease in the etiology of dementia.

Whereas cerebral perfusion and CT-imaging as discussed above provide a measure of large vessel damage, we used retinal imaging since the inception of the Rotterdam Study to directly quantify the small vessels. We found that retinal vascular calibers relate to cerebral atrophy, especially white matter atrophy, and with worse white matter microstructure [57, 58]. In addition, increasing evidence supports an association between retinal vascular changes and dementia, especially Alzheimer’s disease [59].

GENETIC RISK FACTORS: DISENTANGLING THE COMPLEXITY OF DEMENTIA

In addition to modifiable risk factors that accumulate during life, part of the susceptibility to dementia is already determined at conception by your genetic make-up [60]. Estimates for the heritability of dementia syndromes vary greatly, suggesting anything from a predominant genetic component to a minimal influence by genes [61]. While the level of heritability may differ depending on the methodological approach and the specific study population, it is clear that genes do play a role in dementia [62]. The first genetic risk factors for dementia were identified for familial forms by studying affected pedigrees [60]. However, sporadic forms, which are the most common, do not result from one specific mutation but rather they are the consequence of multiple risk increasing variants that may or may not lead to dementia depending on other genetic variants or additional environmental factors. Of these, APOE is the most well-known due to its large effect size in combination with a high frequency of the risk alleles in the general population [63–65].

Many candidate variants have been studied in relation to dementia, but these unfortunately rarely replicated. With the advent of large scale genome-wide association studies, robust associations have been identified between common variants and dementia [62]. Their effect sizes are modest and with odds ratios below 2.0, and generally much smaller [66]. Recently, rare variants with slightly larger effects have been identified through sequencing studies [67]. The participation of the Rotterdam Study in these discoveries was mainly by contributing samples to large consortia such as CHARGE and IGAP [68, 69].

However, the Rotterdam Study has been more actively leading initiatives to map genetic determinants of endophenotypes of dementia. While dementia is a heterogeneous syndrome resulting from a multitude of factors, endophenotypes are thought to represent a more distinct disease process that is closer to the underlying biology. For example, amyloid-β levels in plasma or cerebrospinal fluid are likely to be more specific markers for the amyloid cascade, while measures of the cerebral perfusion may indicate vascular pathways [25]. We have focused our endophenotype studies mainly on neuroimaging markers retrieved from MRI. In the Rotterdam Study, over 13,000 scans have been performed using the exact same MR machine and acquisition protocol, making it the single largest study with such data [36]. The first genetic studies of imaging markers were published in 2012, where we studied hippocampal and intracranial volume [70, 71]. We found that robust association signals could be identified in samples of around 10.000 individuals and that some of these relate to risk of dementia [72, 73]. After these initial publications, larger studies have been performed on these and additional neuroimaging measures [74–76]. We have performed genetic association studies of the volumes of other subcortical structures [77], and found that amygdala volume might be a good endophenotype for Alzheimer’s disease [77]. An important development in these studies is the availability of biobanks such as the United Kingdom (UK) Biobank, which have provided a large resource of valuable data [78, 79]. Beyond the studies of these gross neuroimaging measures, we have also investigated whether novel high-dimensional imaging markers may be more informative for genetic studies. Two of these are the shape of subcortical structures and voxel-based grey matter morphometry [80, 81]. In both of these studies, it was found that there exists substantial regional variation in the heritability of these high-dimensional markers that would have been missed when looking only at the traditional measures that describe the brain roughly [82]. While the next step would logically be to perform genome-wide association studies of all these novel imaging measures, their sheer number poses a practical obstacle. For voxel-based morphometry, for example, there are 1.5 million voxels in the brain for which genome-wide association studies, with 10 million genetic variants, would result in trillions of association tests. This is computationally intensive but also makes for a stringent multiple testing threshold. To overcome these barriers, we have developed a new analytical framework, HASE, that is specifically designed for high-dimensional analyses [83]. The computational time for such a genome-wide brain-wide association study is greatly reduced from several years to only several hours. This is achieved by implementing smarter algorithms for data storage and retrieval, but also by using a novel form of meta-analysis termed partial derivatives meta-analysis [84]. In a proof of principle study within 4000 individuals of the Rotterdam Study, we performed genome-wide association studies of 7000 voxels in the hippocampi and found this to be indeed computationally feasible [83]. Interestingly, the top variant was located in a locus that has been previously associated with hippocampal volume in a much larger sample. Larger genetic studies of high-dimensional imaging markers are now underway. It remains to be seen whether this approach will lead to the identification of more genetic loci and how these are related to dementia.

IMPLICATIONS FOR CURRENT AND FUTURE TRIALS AND THE ROLE OF PREDICTION

With high failure rates of large, phase III trials of disease-modifying or arresting drugs, the search of finding successful pharmacological therapies for this detrimental disease is one the most challenging and expensive healthcare issues to date [85–87]. To combat this challenge, the focus has shifted from development of treatment strategies in advanced disease stages toward preventive intervention approaches in asymptomatic states or early disease to delay or prevent the onset of dementia. As such, pharmaceutical companies, policy makers and trialists are looking at population-based studies to inform them on optimal design of preventive trials. Such contributions are first in the form of mapping the potential for prevention given current knowledge of potentially modifiable risk factors and second providing risk models to identify high-risk individuals that would benefit most from an effective intervention and therefore should primarily be targeted for trials. We discuss both aspects here.

Back in 1996, a first report came out hinting towards the potential prevention of Alzheimer’s disease and dementia [88]. Over the past decades, this approach was substantiated by extensive evidence on the association of vascular risk factors and the risk of dementia [89–93]. The opportunities of primary prevention through modification of vascular and lifestyle factors has been further fueled by accumulating evidence, including initial observations from the Rotterdam Study, that age-specific incidence of dementia is declining in developed countries [7, 95]. These findings have in part been attributed to a better education in early life, and a healthier lifestyle, including an improvement management of vascular risk. Indeed, recent studies have provided quantitative and convincing evidence that up to a third of all dementia cases may be prevented if modifiable risk factors, such as diabetes, smoking, and physical inactivity, were eliminated [14, 15]. The importance of these observations has recently been underscored and anchored by an international committee of experts, consolidating dementia prevention by means of lifestyle improvement as a global yet ambitious strategy to reduce the burden of this disease globally [30]. Building on evidence forthcoming from these observational studies, several large, randomized controlled trials have been conducted assessing the efficacy of multi-domain lifestyle interventions to prevent cognitive decline in community-dwelling individuals [96–99]. So far, most of these trials have been inconclusive, yet the FINGER trial found evidence that these efforts may be more effective in a high-risk population [96]. Future trials are planned to target these interventions at high-risk individuals in the general population, such as the US-POINTER and SINGER trials [100].

The use of biomarkers is increasingly advocated for purpose of risk stratification in selected, high-risk populations (e.g., memory clinics). These biomarkers include amyloid and tau protein levels assessed by cerebrospinal fluid or positron emission tomography (PET), and rare genetic variants with high individual risk. Yet, such approaches cannot be easily translated to the general population for various reasons. As such, there are currently no established risk prediction models for dementia [101]. In a recent effort, we sought to validate currently proposed risk prediction models, but concluded that beyond age these risk models do not meaningfully contribute to risk prediction of dementia [102].

FUTURE PERSPECTIVES

The size and community scope of the Rotterdam Study necessitate striking a fine balance between the drive to include expensive and burdensome investigations versus feasibility of such investigations in a volunteer population. Other considerations involve the structural shortage of research funding and the advent of big data initiatives like the UK Biobank, German National Cohort and the US Precision Medicine Initiative. The latter encourage seeking a next level of detail in formerly considered large populations of ten to twenty thousand individuals. In such an ever-evolving research field, the choice of phenotypes for the study of dementia in the Rotterdam Study will continue to be driven by earlier observations of risk factors and biological changes in the preclinical disease course and facilitated by technological advancements. Brain MRI has been part of the core protocol since 2005, but more recently functional MRI has been added, and we expect the first results of these efforts shortly. Moreover, addition of arterial spin labelling sequence, with the possibility of including a vasomotor challenge, may add the desired detail to map cerebral hemodynamics in the population, as could more novel measurements like near infrared spectroscopy for (changes in) frontal lobe perfusion, and sidestream dark field imaging for direct visualization of the capillaries. Contrast enhancement could now add further insight into blood-brain barrier function [103], which is increasingly recognized to play a role in Alzheimer’s disease [104]. Other brain imaging techniques, including the use of PET amyloid tracers, are still expensive to apply in large numbers of individuals, but become applicable in smaller studies embedded in the larger design of the Rotterdam Study. Mapping trajectories of amyloid deposition with repeated measures in (initially) healthy individuals is vital to determine the disease course, identify why neuropathology accumulates in many individuals, and why this leads to cognitive disturbances in some, but not others. Beyond brain imaging, the Gothenburg studies have shown that cerebrospinal fluid sampling in forthcoming healthy individuals is safe and feasible [105]. This would not merely benefit etiological and prognostic research of amyloid and tau (for which less invasive means could now suffice), but also allow to determine passage of metabolites through the blood-brain barrier, measure intracranial pressure [106], and perhaps most importantly ready research for the development of novel, yet unidentified markers that can be aptly assessed and validated in previously collected community samples. This preparation is crucial to have long-term follow-up of participants available for rapid validation of proposed prognostic markers for dementia, in CSF as well as plasma and serum. Understanding differences and similarities between measures in CSF and plasma by direct comparison will likely be key to development of useful markers in the latter. Notable candidates for such markers involve angiogenesis [107], lipid transport and metabolism [62], and inflammation [108]. Genetic studies invariably implicate immune response in the onset of Alzheimer’s disease [62, 67]. We have previously found support for a role of inflammation in the Rotterdam Study [109], yet repeated measurements of ideally more specific cytokines are needed. Other candidate markers may in the coming years be identified from ongoing collaborative efforts that map metabolomic changes in the periphery [110].

CONCLUSION

In conclusion, for nearly 30 years now the Rotterdam Study has contributed greatly to the understanding of dementia, in terms of incidence, risk factors, pathobiology, and prognosis. It achieved its success through exploring novel underlying pathologies, pioneering various emerging technologies in a population-based setting, and maintaining a methodologically sound basis. At the same time, the Rotterdam Study has been a key contributor to various worldwide collaborations. In coming years, we expect the Rotterdam Study to continue its contribution within the vast landscape of dementia research.