MRI-Based Biomarkers of Accelerated Aging and Dementia Risk in Midlife: How close are we?

Abstract

The global population is aging, leading to an increasing burden of age-related neurodegenerative disease. Efforts to intervene against age-related dementias in older adults have generally proven ineffective. These failures suggest that a lifetime of brain aging may be difficult to reverse once widespread deterioration has occurred. To test interventions in younger populations, biomarkers of brain aging are needed that index subtle signs of accelerated brain deterioration that are part of the putative pathway to dementia. Here I review potential MRI-based biomarkers that could connect midlife brain aging to later life dementia. I survey the literature with three questions in mind, 1) Does the biomarker index age-related changes across the lifespan? 2) Does the biomarker index cognitive ability and cognitive decline? 3) Is the biomarker sensitive to known risk factors for dementia? I find that while there is preliminary support for some midlife MRI-based biomarkers for accelerated aging, the longitudinal research that would best answer these questions is still in its infancy and needs to be further developed. I conclude with suggestions for future research.

1. Introduction

The global population is aging (Lutz et al., 2008) with projections that the number of people over 60 will more than triple by 2050 (UN, 2007). This broad demographic shift will be more extreme in developed countries like the US, Japan, New Zealand and much of Europe (Christensen et al., 2009). These demographic shifts reflect the progress of modern medicine in the last two centuries as well as advances in treating infectious disease (Armstrong et al., 1999), minimizing the risks of childbirth (Hogan et al., 2010), extending lifespans (Crimmins, 2004), and widespread increases in food production (Pingali, 2012). While such progress is evidenced by an aging population, an aging population is also an imminent warning of public health challenges on the horizon (Christensen et al., 2009).

Extra years of life are not necessarily extra years of health. Despite advances in medicine, the global burden of years lost to disability has risen in the last 20 years due to an aging population coupled with a stable rate of age-related disease (Vos et al., 2012). While many organ systems are impacted by aging, deterioration of the brain is a particularly debilitating form of age-related disease. Cognitive decline is an insidious consequence of brain aging that is closely linked to a decline in the ability to perform everyday tasks, maintain independence, and care for oneself (Kawas et al., 2000; Tucker-Drob, 2011). Decline in cognition is a hallmark of a class of disorders known as dementia. Dementia is an umbrella diagnosis that encompasses an array of disorders with heterogenous etiology. While in certain subtypes of dementia, cognitive decline can be due to isolated factors such as infectious disease, acute traumatic brain injury and rare genetic disorders, aging research commonly focuses on dementia trajectories that are categorized as “neurocognitive disorders of aging” (American Psychiatric Association, 2013). This subset of aging-related dementias includes Alzheimer’s disease, frontotemporal dementia and vascular dementia.

Historically, aging-related dementia research has primarily studied older adults. Chronological age is indeed the biggest risk factor for dementia (Kawas et al., 2000; Launer et al., 1999). However, interventions targeting older adults have largely proven to be ineffective at limiting morbidity and disability (Dehnel, 2013). A few recent illustrative examples have shown weak outcomes in randomized control trials aimed at slowing age-related degeneration with the Mediterranean diet (Valls-Pedret et al., 2015), hypertension reduction (Williamson et al., 2014), cardiovascular care (van Charante et al., 2016), and exercise (Sink et al., 2015). These failures are hypothesized to be a consequence of years of irreversible sub-clinical organ decline that accumulates throughout the lifespan (Finch, 2004; Gillman, 2005). If this is true, interventions may be failing to slow age-related disease because they are implemented too late in the aging process after decline has taken hold.

In a newer model of geroscience, rather than being conceptualized as an isolated stage of life, aging is viewed as a lifelong process that results from the slow accrual of damage to organ systems throughout the lifespan (Moffitt et al., 2017). Clinically detectable aging-related dementia and cognitive impairment represent the end of a lifelong aging process in which genetic predisposition (Strittmatter et al., 1993), environmental insult (Reuben et al., 2017), and lifestyle factors (Fratiglioni et al., 2004) interact to form dementia’s etiology. Consistent with this view, many risk factors for dementia are measurable in childhood (Russ et al., 2017) and middle age (Gottesman et al., 2017) and in some cases risk factors, like blood pressure, may even more predictive of later dementia risk when measured in midlife than when measured in old age (Fitzpatrick et al., 2009). This lifespan view of aging, together with the disappointing results of dementia prevention trials in late life, suggest that there is a promising opportunity to design interventions that target individuals at risk for age-related disease in early to midlife, to slow the aging process before debilitating decline in organ systems has been cemented (Sierra, 2020).

While shifting dementia prevention research from older adults to midlife adults represents a promising approach to extend the health span, it also brings along methodological dilemmas. In the prototypical dementia prevention trial, an intervention is given to a group of older adults in their 60s and 70s who are at high-risk for dementia. This group is then tracked for a number of years and their cognitive decline and dementia prevalence is quantified and compared to a control group who did not receive the intervention. In these trials, the measured outcome (dementia prevalence and cognitive decline) and the target of prevention are one and the same. While methodologically sound, this type of intervention design will be challenging to export to interventions targeted at individuals in midlife. For example, if an anti-dementia intervention is administered to individuals in their 40s instead of their 60s, a clinical trial will have to last 20 extra years in order to use dementia onset as an outcome measure because the onset of aging-related dementia is uncommon before the age of 65 (Launer et al., 1999). Adding 20 years of follow-up to the typical intervention trial will drastically increase the financial burden of these studies and severely limit the number of trials that can be funded.

One proposed solution is to find biomarkers that index sub-clinical changes in biology and cognition that are indicative of premorbid dementia and are markers of decline that has yet to be fully realized. This biomarker approach has already begun to find success in quantifying individual differences in midlife aging (Niedernhofer et al., 2017). A multi-panel biomarker of aging related measures has been linked to cognitive decline, deterioration in health and accelerated facial aging in a population representative sample of 38 year-olds (Belsky et al., 2015). In addition, calorie restriction was found to slow a biomarker of biological aging in a randomized control trial (Belsky et al., 2017b). However, while these multi-panel biomarkers of aging hold promise for many intervention trials targeted at midlife, they may not be an ideal biomarker for aging-related dementia studies because they do not directly measure degradation of the brain, the one organ directly responsible for cognition, memory, and intellectual function. In a search for sensitive biomarkers of accelerated midlife aging it is important for measurement to remain close to the putative etiological pathway of neural change that connects midlife aging to dementia in older adults.

Our current understanding of the links between cognitive aging and dementia suggest that subtle changes in brain pathology may be detectable throughout the adulthood and that accurately quantifying these changes in midlife may reveal further insight into dementia risk (Sperling et al., 2014). Therefore, I begin this review with a brief overview of age-related changes in cognition, their relationship to dementia and their implications for MRI biomarkers (section 2). I then review the human MRI literature in a search for potential brain biomarkers that are most likely to index accelerated midlife aging and dementia risk (section 3). My review of the human MRI literature will be separated into two major sub-sections: 1) studies of gray matter and 2) studies of white matter. I focus this review on these structural MRI modalities for several reasons. First, structural MRI has been a primary focus of many studies of development and aging in the literature. Second, structural measures have moderate to high measurement reliability (Han et al., 2006; Iscan et al., 2015; Vollmar et al., 2010) unlike many functional MRI measures (Elliott et al., 2018; Elliott et al., 2019). Third, structural MRI is likely closer to the putative causal pathway linking neurodegeneration to cognitive decline than functional MRI (Wilson et al., 2010). Fourth, structural MRI will be collected in over 4,000 participants of the Human Connectome Project (Van Essen et al., 2013), 10,000 participants in the ABCD study (Volkow et al., 2017), and 100,000 participants in the UK Biobank (Miller et al., 2016), therefore many researchers will model their protocols after these large studies. Fifth, while positron emission tomography (PET) is widely used in quantifying risk for Alzheimer’s in old age (reviewed elsewhere (Ishii, 2014)), it is an invasive measure that requires exposure to radiation and therefore it has not been widely studied in early adulthood and midlife (Jack et al., 2013). For these reasons, a timely review of structural MRI biomarkers could have broad reach in informing future aging research and dementia interventions.

2. Cognitive Aging and Dementia

To understand what a successful midlife biomarker for accelerated brain aging and dementia risk would look like, it is important to first understand what cognitive aging looks like across the lifespan and how cognitive aging is related to the diagnosis of dementia in older adults. For a midlife MRI-based biomarker to be useful in dementia risk stratification and for advancing midlife treatments for dementia, it must first track eiology that links midlife cognition and cognitive decline to dementia in older adults. However, there are several lessons and complexities from cognitive aging and dementia research that will inform the search for MRI biomarkers of accelerated aging and dementia risk.

When measured across the lifespan cognitive abilities show a general pattern of rapid development in childhood before stabilizing in adolescence and declining throughout adulthood (Salthouse, 2004). For example, the general factor or “g factor” of intelligence (Spearman, 1904), which is indexed by full-scale Intelligence Quotient (IQ) tests, has been estimated to negatively correlate at r = −.48 with age in adulthood (Deary et al., 2010). While decline in many cognitive abilities like reasoning, perceptual speed, and episodic memory begin in the early 20s, this trajectory is not universal to all cognitive domains. For example, the size of an individual’s vocabulary continues to grow throughout much of adulthood, correlating at r = .63 with age (Deary et al., 2010) and peaking at around 55 years of age before beginning to decline (Hartshorne and Germine, 2015). While these general trends of normative cognitive aging are well documented, there are several difficulties in interpreting relationships between age-related cognitive decline, peak cognitive ability, dementia risk factors and dementia diagnosis.

One challenge is illustrated in the observation that individuals with higher childhood IQs and greater educational attainment are less likely to be diagnosed with dementia in old age (Baumgart et al., 2015; McGurn et al., 2008; Whalley et al., 2000). However, higher education attainment is not associated with a slower rate of cognitive decline in adulthood (Seblova et al., 2019; Zahodne et al., 2011). Further substantiation of this account comes from natural experiments, which have found that additional years of education may causally raise overall level of cognitive ability in early adulthood (Ritchie and Tucker-Drob, 2018) and from meta-analytical evidence that suggests that peak cognitive ability, in addition to educational attainment, has no association with the rate of age-related cognitive decline (Tucker-Drob et al., 2019). While some theories of aging and dementia have suggested that the that education reduces the risk for dementia by slowing the rate of cognitive decline (Stern, 2012), this growing body of evidence indicates this may not the case (Tucker-Drob, 2019). Thus, while educational attainment is associated with where cognitive ability peaks, it does not seem to be associated with the rate at which it declines.

Another challenge to interpreting dementia risk factors is due to limitations intrinsic in our current methods for diagnosing dementia. To qualify for the DSM-V diagnosis of dementia, an individual must have substantial cognitive impairment that interferes with activities of daily living. This impairment must be the result of a decline from a higher level of functioning and this decline must be in excess of what is expected in normal aging (American Psychiatric Association, 2013). While clear conceptually, these criteria suffer from several limitations when applied in practice. Let us first consider the criteria of cognitive impairment. Cognitive impairment is typically measured with a short neuropsychological assessment (e.g. Mini-Mental State Examination (MMSE)). These short assessments have been designed to discriminate individuals with normal range cognitive functioning from those with cognitive impairment by employing a numerical threshold for determining dementia status (Tucker-Drob, 2019). While this approach may be useful for measuring current cognitive impairment, when used alone, a single MMSE score cannot distinguish between individuals with a low cognitive peak and a shallow decline from those with a high cognitive peak and steep cognitive decline (see figure 2). While clinicians and researchers attempt to overcome this limitation by assessing “decline from a previous level of higher functioning”, rigorous, longitudinal measurements of cognitive function are rarely available and instead cognitive decline is typically assessed through self-report and/or informant-report from a family member. Unfortunately, human memory and self-report are limited and can lead to imperfect assessment of the extent and rate of cognitive decline.

Figure 2.

This figure represents hypothetical trajectories of cognitive function across the lifetime of two hypothetical individuals. The individual in dark purple reaches higher peak level of cognitive ability than the individual in light purple. The dotted lines represent a theoretical trajectory in which each individual’s peak cognitive is maintained throughout adulthood. The solid lines represent “normal” cognitive aging trajectories in which each individual reaches peak cognitive functioning in early adulthood before experiencing mild cognitive decline into old age. The dashed lines represent an accelerated trajectory of cognitive decline in adulthood that is indicative of pathological aging that would result in a dementia diagnosis for both individuals. Of note, under some dementia screening regimes, including many that rely on the short cognitive tests like the mini mental status exam, the solid light purple trajectory could result in a dementia diagnosis due to a drop below a threshold of cognitive function. However, in this case, the person represented by the solid light purple line has experienced normative cognitive aging and the diagnosis is attributable to a low-level of peak cognitive function, not an accelerated rate of cognitive decline. This possibility illustrates a potential limitation in our current method for diagnosing dementia. Of note, for simplicity, this figure assumes a categorical model of dementia whereby the rate of cognitive decline either has a “normal trajectory” (solid line) or a “pathological trajectory” (dashed line) (Tucker-Drob, 2019). Alternatively, under a continuous model, across-individuals there would be continuous variation in cognitive aging trajectories between the dotted and dashed line. The distinction between a “normal” and a “pathological” trajectory would therefore be a matter of degree instead of kind.

Together these practical limitations may result in over-diagnosing dementia in those with low baseline levels of cognitive function who have normal levels of aging-related cognitive decline, while under-diagnosing those with high peak levels of cognitive functioning and an accelerated rate of cognitive decline (see figure 2). This happens because individuals with lower peak cognitive ability are closer to a diagnostic threshold for cognitive impairment and thus are more likely to drop under that threshold due to age-related cognitive decline. This potential bias in diagnosis may help explain the association between education and dementia risk despite the lack of association between education and cognitive decline – education and peak cognitive ability are associated and dictate the distance that cognitive ability has to drop before dipping below a diagnostic threshold. These limitations in diagnosis have led some researchers to call for a change-based assessment of dementia, so that neuropsychological assessment can more accurately identify individuals with cognitive decline that is in excess of “normal aging” (Tucker-Drob, 2019). For an MRI biomarker to be an effective index of accelerated brain aging and dementia risk in midlife, it should be a marker of accelerated decline in brain integrity that is unconfounded by overall level of cognitive function in early adulthood. Throughout this review, I will leverage results from cross-sectional and longitudinal research to evaluate the extent to which current MRI-biomarkers live up to this aspirational goal.

3. Candidate MRI-Based Biomarkers of the Aging Brain

For each MRI measure considered here, I will review recent literature to determine the potential of each as a biomarker for measuring accelerated brain aging and stratifying dementia risk in midlife. Specifically, in discussing each modality I will consider three complementary features of an MRI-based biomarker for midlife brain aging and dementia risk. First, I will investigate changes across the lifespan. To fully understand what accelerated aging looks like, we must first identify the normative age-related pattern of change. A biomarker sensitive to accelerated midlife aging should first demonstrate biologically meaningful age-related variance. Second, I will review the relationship between the proposed MRI biomarker and cognition throughout the lifespan. I will review associations with both cognitive ability and cognitive decline, as they represent dissociable features of dementia risk. As discussed above (see section 2), the ideal MRI biomarker would index cognitive decline over and above cognitive ability. Lastly, if a biomarker indexes brain changes that put individuals at increased midlife risk for the later onset of dementia, it should be linked to known midlife risk factors for dementia that are likely to accelerate the rate of age-related decline including hypertension, obesity, diabetes, traumatic brain injury and genetic risk for dementia (e.g. APOE status) (Baumgart et al., 2015; Liu et al., 2013). If aging is truly a lifetime process and risk for dementia is increased by a lifetime burden of exposures, then associations between risk factors and MRI-based biomarkers may be detectable before the onset of clinically diagnosable dementia.

While I review biomarker associations across the lifespan, I will prioritize evidence from several developmental periods where biomarkers associations may provide the most illuminating information about a biomarker’s potential as an indicator of accelerated aging and future dementia risk. First, due to the centrality of cognitive decline to dementia, associations between a biomarker and the earliest signs of cognitive decline in midlife, years before dementia is clinically diagnosed, will be particularly informative. Second, while associations with cognitive decline will be emphasized, there are many more studies of MRI associations with cross-sectional cognitive ability. In these cases, it is less important for a biomarker to be associated with cognitive ability in childhood or midlife, as individual differences in early life cognitive ability are primarily driven by peak cognitive ability instead of cognitive decline. In contrast, associations between a biomarker and cognitive ability in older adults will be more indicative of the accumulated impact of cognitive decline (Cox et al., 2019). Lastly, associations between a biomarker and dementia risk factors in midlife, rather than late life, may be most indicative of the modifiable impact of dementia risk factors that have begun to accrue before dementia has set in (Abell et al., 2018; Launer et al., 2000; Qiu et al., 2005). An MRI measure with a strong evidence base of associations with age-related decline across the lifespan, cognitive decline and dementia risk factors in midlife and cognitive ability in late life would have further credence in favor of being a viable biomarker for midlife dementia risk.

3.1. Gray Matter Measures

Historically, the most common MRI method for the measurement of gray matter has been voxel based morphometry (VBM), which is a measurement of the local density of gray matter (Ashburner and Friston, 2000). Although VBM has been widely used to quantify age-related changes in gray matter (Good et al., 2001; Sowell et al., 2004, 2003) as well as group differences in age-related pathology (Ferreira et al., 2011), VBM is made up of two distinct components: cortical thickness and surface area (Panizzon et al., 2009). Cortical thickness represents the distance from the outer white matter boundary to the outer edge of the pial surface and is driven by the density of cells in a cortical column, while surface area represents the 2-dimensional area of the gray matter surface and is modulated by the number of cortical columns in a region (Rakic, 1988). Cortical thickness and surface area have been shown to be genetically and phenotypically independent (Panizzon et al., 2009; Winkler et al., 2010, 2009), calling into question the wisdom of measuring their combination through volume metrics like VBM instead of measuring both surface area and cortical thickness separately (Winkler et al., 2018). The difference between these measures may be particularly important in the study of age-related change because cortical thickness and surface area have been shown to have independent developmental trajectories (Hogstrom et al., 2013; Lemaitre et al., 2012; Wierenga et al., 2014). Furthermore, individual differences in gray matter volume are largely redundant with individual differences in cortical thickness and surface area. Cross-sectional gray-matter volume is highly correlated with surface area across subjects (r > 0.8) and only weakly correlated with cortical thickness (Winkler et al., 2010). Alternatively, longitudinal age-related decline in gray matter volume is largely driven by cortical thinning (r > .9) and weakly anticorrelated with surface area changes (Storsve et al., 2014). Therefore, in this review of candidate biomarkers for midlife aging, I will focus the primary review on studies that report cortical thickness (Figure 2A) or surface area (Figure 2B) and review the VBM literature when it is a useful source of complementary information.

3.1.1. Lifespan Developmental Changes

Substantial evidence now suggests that gray matter change is the rule, instead of the exception, throughout the human life course (Hogstrom et al., 2013; Potvin et al., 2017; Shaw et al., 2008). While there is some disagreement about whether cortical thickness in the brain peaks around age 2 (Li et al., 2013; Lyall et al., 2015) or later on in childhood (Giedd et al., 1999; Lenroot and Giedd, 2006; Shaw et al., 2008), large cross-sectional studies provide evidence that across the vast majority of human development the cortex is thinning (Potvin et al., 2017). This general pattern appears to play out in surface area as well, with some small caveats (see figure 3). While cortical thickness reaches its peak in young childhood, surface area continues to grow throughout adolescence (Wierenga et al., 2014), accounting for most of growth in cortical volume after age 2 (Lyall et al., 2015). However, around age of 20 surface area is no longer expanding and shows a consistent pattern of decline into old age (Hogstrom et al., 2013; Lemaitre et al., 2012; Potvin et al., 2017; Storsve et al., 2014).

Figure 3.

Visual approximate representation of the relationship between age and each of the MRI biomarkers reviewed. These trajectories are extrapolated from the cross-sectional and longitudinal studies surveyed in this review.

While cortical thickness and surface area both show patterns of decline after the age of 20, the rates of decline vary, are non-linear across the lifespan and heterogeneous across brain regions (Fjell et al., 2013). Large cross-sectional imaging studies have demonstrated that age is nonlinearly related to cortical thickness across the lifespan (McKay et al., 2014). Cross-sectional and longitudinal studies have both found that nonlinearities appear are manifest as pronounced declines of nearly 1% per year in cortical thickness before the age of 14 and after the age of 60 (Fjell et al., 2013, 2015a; Potvin et al., 2017). In midlife there appears to be a linear and stable rate of decline in cortical thickness of between .1 and .5% per year (Fjell et al., 2013). Longitudinal data following subjects for 1–7 years suggest that the pattern of decline in surface area appears to be more consistent and linear with an average rate of decline after the age of 20 of approximately .4% per year (Schnack et al., 2015; Walhovd et al., 2016). However, constant gray matter decline is not universal across the brain. The entorhinal cortex, a region crucial for memory formation (Fyhn et al., 2004) and central to age-related changes in memory ability (Fjell et al., 2014b; Killiany et al., 2002), appears to be one notable exception to the rule. Bucking the pattern of consistent decline after 20 years of age, a large cross-sectional study (N = 1,660) suggests that entorhinal cortex thickness peaks at age 44 before beginning to decline (Hasan et al., 2016).

Overall, regional patterns of decline appear to be non-random, and instead follow developmental and evolutionary trends. Multiple studies have demonstrated that frontal and parietal cortical regions are amongst the slowest regions to develop and the fastest to decline (Hill et al., 2010; Lemaitre et al., 2012; Lenroot and Giedd, 2006; Potvin et al., 2017; Wierenga et al., 2014). These regions have the most complex cortical architecture (Fjell et al., 2014a), overwhelmingly support higher-order cognitive processes, and are transmodal, sitting on top of the sensory processing hierarchy where they bind information from unimodal regions (Margulies et al., 2016; Mesulam, 1998). These observations have been synthesized as the “last in, first out” theory of gray matter cortical development (Douaud et al., 2014; Raz, 2000; Salat et al., 2004). This theory suggests that regions supporting the most complex aspects of human cognition take the longest to develop and are also hit the earliest and hardest by aging. These frontal and parietal brain regions also show the most cortical expansion over evolutionary time and are the most synaptically and dendritically complex (Buckner and Krienen, 2013; Fjell et al., 2015b; Hill et al., 2010).

3.1.2. Covariation with Cognition and Cognitive Decline

Beginning in childhood, both cortical thickness and surface area are linked to individual differences in cognitive ability. Children with thinner cortices perform better on tests of memory, visuospatial functioning, spatial reasoning, and problem solving (Squeglia et al., 2013). However, the relationship between thinner cortex and cognitive ability is not stable across the lifespan. Longitudinal analyses that have tracked individuals for 3–5 years at different ages suggest that in adulthood the pattern reverses and that individuals with higher cognitive ability tend to have thicker cortices (Schnack et al., 2015). Midlife appears to be an important window during which higher cognitive ability is no longer associated with refinement and pruning of cortex, through thinning, and instead maintenance of cortical thickness becomes important for maintaining cognitive ability. In contrast to cortical thickness, larger surface area is stably associated with higher cognitive ability across the lifespan (Walhovd et al., 2016). Longitudinal measurements reveal that across adulthood and into old age cortical surface area declines in parallel trajectories for individuals with high and low cognitive ability (Walhovd et al., 2016).

While some of the studies linking gray matter development to cognitive ability have longitudinal MRI data and large sample sizes, nearly all lack repeated measurements of cognitive function that could be used to control for premorbid cognitive ability and measure change in cognitive ability within the same individuals. Longitudinal measurements of cognition are important because they allow researchers to disassociate peak cognitive ability from the rate of cognitive decline. In a unique sample, the Lothian Birth Cohort, cognitive ability was measured in the same individuals at age 11 and 73. Using this sample researchers have found that cognitive decline over the course of 60 years is related to smaller surface area throughout the brain but seems to have no association with cortical thickness (Cox et al., 2017; Karama et al., 2014). Notably, many of the regions that demonstrate the largest associations with cognitive decline are the same frontal and parietal regions where surface area shows the latest maturation in development and the strongest associations with cognitive ability (Deary et al., 2010; Schnack et al., 2015). The specificity of surface area in these findings is also consistent with a large twin study demonstrating that cognitive ability is phenotypically and genetically associated with surface area but not cortical thickness (Vuoksimaa et al., 2015). In addition, using the Lothian Birth Cohort, researchers also found that cognitive decline in later life (age 74–77) was strongly associated with the longitudinal atrophy of gray matter volume (Cox et al., 2020).

3.1.3. Associations with Dementia Risk Factors

There are a number of modifiable dementia risk factors extrinsic to the brain. An influential recent review of this literature found support for the following phenotypes as modifiable risk factors of cognitive decline and dementia: traumatic brain injury, midlife obesity, midlife hypertension, smoking, and diabetes (Baumgart et al., 2015). Importantly, these risk factors can all be assessed throughout an individual’s lifetime, are predictive of later dementia risk, and are modifiable targets for interventions in midlife. Therefore, if MRI measurements of gray matter are to be biomarkers for dementia risk, they should also be sensitive to exposure to these modifiable risk factors and may be reversible if the risk factors are mitigated.

It is important to note that the modifiable risk factors for dementia are not all independent. A correlated cluster of risk phenotypes has been labeled ‘vascular risk’ (Gottesman et al., 2017) and is made up of physical inactivity, poor diet, obesity, hypertension, smoking and diabetes (Gorelick et al., 2011; Iadecola, 2013; Kivipelto et al., 2001; O’Brien and Thomas, 2015). In observational studies, vascular risk has been broadly associated with a reduction in gray matter (Tchistiakova and MacIntosh, 2016). Furthermore, hypertension in older adults has been linked with decreased overall cortical thickness, driven by thinner frontal, parietal, and temporal cortices (Alosco et al., 2014). This association between hypertension and a widespread pattern of thinning, as well as brain volume reduction, has been corroborated by other studies (Cardenas et al., 2012; Leritz et al., 2011; Seo et al., 2012) and has been shown to be partially independent of cholesterol and amyloid-beta, a peptide that forms plaques in Alzheimer’s disease (Villeneuve et al., 2014). In addition, a large, recently conducted study in the UK Biobank found converging evidence of associations between multiple risk vascular risk factors (including diabetes, hypertension, body mass index and waist-hip ratio) and lower gray matter volume, with the largest effect sizes in frontal and temporal regions (Cox et al., 2019). However, in most studies these effects have been limited to measurements of hypertension and cortical thickness taken at the same time point in older adults. In one study with longitudinal assessment of hypertension, a similar pattern of cortical thinning in old age, this time in insular, frontal, and temporal cortices, was found to be linked to midlife hypertension measured 28 years before cortical thickness was assessed (Vuorinen et al., 2013). Together these results suggest that hypertension across the lifespan is linked to cortical thinning.

In general, this accelerated pattern of cortical thinning and atrophy is present for several dementia risk factors. Cortical thinning has been found in studies of obesity (Hassenstab et al., 2012; Marqués-Iturria et al., 2013), body mass index (Veit et al., 2014), diabetes (Mosconi et al., 2018; van Velsen et al., 2013), smoking (Karama et al., 2015; Sutherland et al., 2016), and traumatic brain injury (Hayes et al., 2017; Stillman, 2008). Complementing these findings, cortical sparing (i.e., decreased rate of thinning with aging), has been linked to higher levels of physical activity (Erickson et al., 2013; Walhovd et al., 2014), cognitive training (Engvig et al., 2010) (but see (Thomas and Baker, 2013) for a critical review), as well as increased adherence to a Mediterranean diet (Mosconi et al., 2018; Staubo et al., 2017). Again, this thinning has been found to be most prominent in frontal, temporal, and parietal regions supporting higher-order cognition, which are also most prone to neurodegeneration associated with dementia and cognitive decline (Fjell et al., 2014a). Furthermore, while several large cross sectional studies have failed to find between APOE genetic risk and brain volume or cortical thickness (M. Habes et al., 2016; Lyall et al., 2019), longitudinal research found that higher APOE risk for dementia was associated with a faster rate of cortical volume reduction and thinning of the hippocampus that was detectable in midlife (Mishra et al., 2018).

Given that most of these studies are based on cross-sectional associations these findings must be interpreted with caution. It is important to disentangle the relative contributions of potential causal directions, as it is often unclear to what extent these associations are driven by healthy brains choosing healthier behaviors or healthier behaviors leading to healthier brains (Belsky et al., 2015). For example, people with thinner cortex may have lower cognitive ability and make unhealthier choices and develop hypertension at a higher rate rather than hypertension causing a reduction in cortical thickness. Cross-sectional studies alone cannot determine causal direction. More longitudinal studies linking change in dementia risk factors to change in the gray matter of the brain are needed to determine the relative contributions of causal influences to these cross-sectional associations. Furthermore, randomized control trials of interventions targeting these dementia risk factors with longitudinal measurement of the brain are needed to best determine causal relationships between dementia risk factors, cortical thickness and surface area.

In summary, cortical thickness and surface area each partially fulfill the 3 features of a midlife biomarker for dementia risk. First, surface area and cortical thickness show consistent patterns of change throughout development and aging. The pattern of decline is nonuniform across the cortex and there is evidence that evolutionarily expanded brain regions of cortex supporting higher-order cognitive functions may be most susceptible to age-related neural degradation. Second, cross-sectional data suggests that, in the second half of the lifespan, both thicker cortex and larger surface area is associated with cognitive ability. Third, there are signs that surface area and cortical thickness are linked to dementia risk factors. While these general trends are clear in the literature, the vast majority of the current evidence for aging in the brain comes from cross-sectional data. Cross-sectional studies can suggest links between age and cortical thinning or reduced surface area but are ultimately limited in their ability to infer within-person change. Future longitudinal research with multiple measurements of cognitive ability, dementia risk factors and gray matter integrity will be critical to understand the interplay between cognitive beginnings, cognitive decline and brain aging.

3.2. Measures of White Matter

Since the late 1980s, one of the most commonly used MRI methods for identifying age-related changes in white matter has been fluid attenuated inversion recovery (FLAIR) imaging (Wardlaw et al., 2015). FLAIR imaging can detect white matter hyperintensities in the brain (Figure 2D) that indicate relative increases in water content and water mobility in white matter (Wardlaw et al., 2015). These changes in water content have been shown to arise from a heterogenous array of tissue damage including microbleeds, stroke, hemorrhages, demyelination, and axonal loss (Gouw et al., 2011).

As MRI technology has developed, diffusion tensor imaging (DTI) has emerged as a complementary technique to FLAIR imaging. DTI is capable of measuring more subtle variation in white matter degradation and is sensitive to individual differences in white matter integrity as well as the pathological degradation represented by white matter hyperintensities (Maillard et al., 2013; Schmidt et al., 2016). The most common DTI metrics used to measure age-related change in white matter integrity are fractional anisotropy (FA) and mean diffusivity (MD). FA measures the directional diffusion of water molecules (Basser, 1995) and ranges from 0 to 1 with higher values representing increased directionality of diffusion (Figure 2C). Without constraints, water molecules will diffuse evenly in every direction, which leads to low FA. Therefore, FA is higher when water diffusion in white matter is constrained by axonal bundles, higher myelin content, and uniform fiber alignment as is found in the healthy brain (Le Bihan, 2003). MD measures the average diffusion of water in all directions. Therefore, lower MD indicates less water diffusion and more tissue constraints characteristic of a healthy brain. In this way FA and MD are complementary, negatively correlated measures that are sensitive to changes in axonal integrity, interstitial fluid mobility, and water content (Basser and Pierpaoli, 1996), and predict the conversion of white matter tissues to white matter hyperintensities (Maillard et al., 2014, 2013). Of note, DTI measures also have several limitations that are important to consider when interpreting associations with aging phenotypes, including potential confounds due to motion, multiple contributors to diffusion differences (e.g. myelination, axon density and membrane permeability) and high sensitivity to the data quality (Jones et al., 2013). In the next section I will focus my review on the biomarker potential of white matter hyperintensities, FA, and MD.

3.2.1. Lifespan Developmental Changes

White matter hyperintensities are typically measured in participants in midlife or later because they are uncommon before the age of 30 (Mohamad Habes et al., 2016). However, after the age of 65 white matter hyperintensities can be found in more than 90% of individuals (Schmidt et al., 2011). A robust finding across studies is a positive association between the overall volume of white matter hyperintensities and age (Van Dijk et al., 2008). This association is log-linear indicating that as people get older, they tend to accumulate white matter hyperintensities at a faster rate (Figure 3). Large cross-sectional studies with thousands of individuals suggests that age explains nearly 25% of the variance in white matter hyperintensity volume (Alfaro-Almagro et al., 2017; Mohamad Habes et al., 2016; Ryu et al., 2014). Furthermore, longitudinal studies have found that as individuals age the accumulation of white matter hyperintensities do not occur randomly but instead reflect the expansion of previous hyperintensities (Duering et al., 2013; Maillard et al., 2012a; Schmidt et al., 2016).

Complementary associations with age across the lifespan have been found with FA and MD. Both cross-sectional and longitudinal studies have shown 5–7% per year increases in FA and decreases in MD from birth to young adulthood (Lebel et al., 2012; Lebel and Beaulieu, 2011). This pattern is consistent with continual myelination and increasing axonal density that is partially responsible for the comparatively prolonged brain development of humans (Lebel and Beaulieu, 2011). Within this pattern of white matter development there is tract-specific heterogeneity. Specifically, white matter tracts that connect frontal brain regions, responsible for the integration of information and complex cognitive tasks, show delayed maturation, while the white matter tracts supporting basic sensory processing are the quickest to mature (Lebel and Beaulieu, 2011). However, in general regional specificity of white matter associations are much less consistent than findings with gray matter.

After childhood, both cross sectional and longitudinal studies have found strong associations between whole-brain decreases in FA and increases in MD with aging (Cox et al., 2016; Sexton et al., 2014). Between the ages of 20 and 40 the development of FA and MD in white matter tracts stabilizes and begins to show the first signs of aging (Lebel et al., 2012). After the age of 40, white matter degradation appears to be nearly universal and the rate of decline increases as individuals age. At the age of 40 decline in white matter integrity is on average less than 1% per year but accelerates to nearly 3% per year by the age of 80 (Sexton et al., 2014).

In old age, DTI measures and white matter hyperintensities are closely related, complementary elements of white matter degradation. FA is lower and MD is higher in white matter hyperintensities as well as the white matter immediately surrounding hyperintensities (De Groot et al., 2013; Maillard et al., 2014). In addition, FA, MD and white matter hyperintensities themselves predict the emergence and spread of additional white matter hyperintensities (Maillard et al., 2013; Wardlaw et al., 2015). While this coupling between subtle changes in white matter and later white matter disease has been shown extensively in older adults it has yet to be thoroughly studied in younger individuals whose brains may be most responsive to anti-aging interventions (Belsky et al., 2017a; Moffitt et al., 2017). Overall, white matter hyperintensities, FA, and MD exhibit change across the lifespan, meeting the first feature of a biomarker for accelerated brain aging and dementia risk.

3.2.2. Covariation with Cognition and Cognitive Decline

The association between white matter hyperintensities and cognition is a robust and well replicated finding (Debette and Markus, 2010; Prins et al., 2005; Vermeer et al., 2003). In large cross-sectional studies, white matter hyperintensities have been associated with lower levels of cognitive ability in older adults (Cox et al., 2019). Meanwhile, longitudinal imaging studies have demonstrated that the progression of white matter hyperintensities coincide with faster rates of cognitive decline in the general population (De Groot et al., 2002; Garde et al., 2005; Prins et al., 2005; Van Dijk et al., 2008) as well as in Alzheimer’s disease (Carmichael et al., 2010). While many studies have linked cognitive ability with white matter hyperintensities in older adults, few have been able to control for childhood cognitive ability. However, results from two studies able to measure cognition longitudinally have demonstrated that the volume of white matter hyperintensities predicts decline in cognitive ability from childhood to midlife (d’Arbeloff et al., 2019) and cognitive decline in older adults after controlling for childhood cognitive ability (Valdés Hernández et al., 2013).

DTI measures of white matter integrity (i.e., high FA and low MD) have also been associated with cognitive ability. While results are more mixed in childhood than adulthood, there is a general trend in which greater white matter integrity across the brain is associated with higher cognitive ability (Haász et al., 2013; Muetzel et al., 2015; Penke et al., 2012, 2010). From the beginning of life through adolescence, white matter integrity is positively correlated with cognitive ability (Lee et al., 2017; Muetzel et al., 2015). This same association between high white matter integrity and high cognitive ability has been consistently found in samples of midlife and older adults (Cox et al., 2019; Madden et al., 2012). While there are studies that show tract-specific associations between white matter integrity and cognitive ability, these findings are heterogenous and have yet to describe a consistent pattern of tract specificity. Future research with larger samples is needed to better understand tract-specific (i.e., regional) associations between white matter integrity and cognitive ability.

An important limitation in many of these studies is that they are cross-sectional with broad age-ranges. Because fluid cognitive ability (see section 2) and white matter integrity (see section 3.2.1) both decline with age, longitudinal studies are needed to test if white matter integrity mediates the association between age and cognitive ability instead of a possible alternative explanation, namely that white matter integrity and cognitive decline are independently associated with age, and their apparent association is driven by age-related confounds in cross-sectional studies (Bennett and Madden, 2014; Madden et al., 2012; Salthouse, 2011). When age has been controlled for in cross-sectional studies, the relationship between white matter integrity and cognitive decline is attenuated, often by 50% or greater, but is still present (Madden et al., 2012). A large, cross-sectional study in the UK Biobank found that a composite measure of both white matter integrity (including FA, MD and white matter hyperintensities) and gray matter volume accounted for more than double the variance in cognitive ability in older adults than middle aged adults (Cox et al., 2019). Evidence from an older cohort further suggests the cross-sectional association between cognitive and brain integrity may continue to strengthen into the eighth decade of life (Ritchie et al., 2015). Together these results provide cross-sectional evidence that these measures may index the effects of cognitive decline that accumulate with aging. Furthermore, evidence from longitudinal studies is also consistent with a model in which white matter integrity tracks cognitive ability (Lövdén et al., 2014; Ritchie et al., 2015). Together these findings implicate lower FA and higher MD as important indicators of white matter degradation that contribute to cognitive decline due to structural “disconnection” in the aging brain (O’Sullivan et al., 2001; Salat, 2011). Thus, while current data provide evidence that white matter hyperintensities, FA, and MD covary with cognition, more longitudinal evidence across the lifespan is needed to more directly link measures of white matter integrity to cognitive decline (Salthouse, 2011).

3.2.3. Association with Dementia Risk Factors

White matter integrity has been linked to many of the same dementia risk factors as gray matter (see section 2.2.3). The strongest links between dementia risk and white matter degradation, including reduced FA, increased MD and increased white matter hyperintensity volume, are with vascular risk factors. This link has been prominently explored in the literature because vascular disease strains cerebral blood vessels, eventually causing small brain bleeds that lead to brain pathology in the form of infarcts, oligodendrocyte damage, demyelination, and lesions (Iadecola, 2013; Wardlaw et al., 2014). Studies investigating specific vascular risk factors have found associations between white matter degradation and hypertension (de Leeuw et al., 2002; Maillard et al., 2012b; van Dijk et al., 2004), diabetes (Ferguson et al., 2003; Gouw et al., 2008; Kodl et al., 2008), high cholesterol (Ryu et al., 2014), obesity (Gustafson et al., 2004; Stanek et al., 2011), and smoking (Gons et al., 2011; Van Dijk et al., 2008). In addition, a large recently conducted study in the UK Biobank found converging evidence across vascular risk factors for an association between vascular risk and lower FA, higher MD and larger white matter hyperintensity volume (Cox et al., 2019). Together these findings are consistent with the hypothesis that chronic vascular strain on the brain leads to age-related white matter pathology (Wardlaw et al., 2013).

White matter degradation has also been linked to non-vascular risk factors like traumatic brain injury (Kraus et al., 2007; Niogi et al., 2008). In addition, APOE genetic risk has been associated with increased white matter hyperintensity volume in mid and late life (Lyall et al., 2019; Rojas et al., 2018). However, in a similar manner to associations with gray matter, large studies have failed to find cross-sectional associations between APOE genetic risk with FA and MD (Lyall et al., 2019). Though, in longitudinal studies APOE risk status has been linked to faster rates of white matter hyperintensity grow, as well as FA and MD degradation (Sudre et al., 2017; Williams et al., 2019). Further, white matter integrity has been linked to factors that are protective against dementia including exercise (Camicioli and Verghese, 2015; Fleischman et al., 2015; Gow et al., 2012) and adherence to a Mediterranean diet (Gu et al., 2016). That said, the evidence base linking these non-vascular and protective factors to white matter integrity is much less extensive.

Because of their long-standing clinical application, white matter hyperintensities have even been used as biomarkers in a number of dementia prevention trials in older adults. These randomized controlled trials have tested interventions intended to reduce modifiable dementia risk factors and measured the accumulation of white matter hyperintensities as a surrogate outcome. Overall, like other intervention studies aimed at preventing dementia, results of these clinical trials have been mixed. A few small studies have found that vascular care and antihypertensive treatment have reduced the progression of white matter hyperintensities (Dufouil et al., 2005; Richard et al., 2010). However, other larger studies have found that hypertensive and cholesterol treatment did not reduce progression of white matter hyperintensities (Ten Dam et al., 2005; Weber et al., 2012). Additional trials targeting vascular risk factors with medication, exercise, and diet are ongoing (Cyarto et al., 2012), but like prior trials these all target individuals in their 60’s, 70’s and 80’s, and may be missing out on the full potential of the biomarker design that could benefit from targeting younger adults who have yet to experience as much age-related decline of the brain. Unlike traditional anti-dementia interventions that use dementia prevalence and cognitive impairment as outcome measures, the use of white matter hyperintensities (or other imaging biomarkers) as outcome measures may allow clinical trials the flexibility to target individuals in midlife who have yet to accumulate decades of age-related decline. Future research should expand the scope of anti-dementia treatment trials to include midlife individuals whose brains are healthier and may be more responsive to intervention.

4. Conclusion and Future Directions

4.1. Summary and Integration: What do we know?

Aging is a lifelong process with a dynamic relationship to changes in cognition. In adults, aging is generally associated with cognitive decline and deterioration of brain tissue. Age-related dementia represents one endpoint of a lifetime of brain aging that ultimately leads to impairments in everyday functioning. Slowing the pace of cognitive aging and development of dementia through midlife preventative efforts are likely to be more effective than the recovery of lost functioning through late life intervention. However, sensitive biomarkers of accelerated aging in the midlife brain are needed to accelerate research and test the effectiveness of earlier interventions (Moskalev et al., 2016).

Overall, MRI measures of gray and white matter integrity show promise as early biomarkers for dementia risk, but further research is needed. Together, cortical thickness, surface area, fractional anisotropy, mean diffusivity and white matter hyperintensities show substantial age-related decline that is continuous between midlife and old age. To varying degrees, they each index individual differences in cognitive ability and cognitive decline, and individual differences in the trajectory of gray and white matter aging are associated with known risk factors for dementia that are measurable in midlife. These MRI biomarkers associations also showed promising developmental patterns. There is initial evidence that some MRI biomarkers are associated with cognitive decline in midlife (d’Arbeloff et al., 2019), both white matter and gray matter associations with cognitive ability strengthen as individuals got older (Cox et al., 2019) and, in the case of APOE, white and gray matter associations track longitudinal decline more strongly than cross-sectional integrity (a pattern parallel to APOE’s association with cognition (Davies et al., 2015; Ritchie et al., 2019)). While recent progress has been made in developing better MRI measurement tools and collecting larger samples, the vast majority of the evidence is still cross-sectional. The lack of large, long-term longitudinal MRI studies of aging and dementia limits our ability to answer important questions about the etiology of dementia and accelerated aging.

4.2. Future Directions for MRI-Based Biomarker Research

A particularly relevant and ongoing debate (Tucker-Drob, 2019) revolves around the extent to which the etiology of cognitive decline in normal cognitive aging and cognitive decline in dementia are shared (i.e., continuous model) or distinct (i.e., categorical model). If age-related dementia and normal age-related cognitive decline have shared etiology, then age-related dementia can be thought of as an extreme version of the age-related cognitive and neurological decline that are experienced by the general population. Alternatively, in a categorical model, while the general population experiences “normal” age-related decline, a subset of the population has distinct etiology that is separately contributing to cognitive decline and it is this unique dementia-related etiology that is accelerating the rate of cognitive decline. One example, that is often used to illustrate the possibility of a categorical model, is autosomal dominant Alzheimer’s disease. In autosomal dominant Alzheimer’s disease, symptoms emerge earlier in life and are attributable to specific genetic mechanisms (Bateman et al., 2011). However, even in autosomal dominant Alzheimer’s disease, symptom onset is still uncommon before the age of 40, suggesting that age-related decline sets the stage for the disease process (Ryman et al., 2014; Sierra, 2020). While the sections of this review, that are focused on connecting aging to dementia, may have reduced relevance if the categorical model of dementia is strictly accurate, a thorough understanding of age-related decline of the brain will still be useful for advancing MRI-based biomarker research because the aging brain sets the stage for both distinct mechanisms of pathological decline and general features of accelerated aging.

Another question critical for the advancement of any measure as a risk biomarker is effect size. Yes, MRI-based measures are linked to age, cognitive ability, and dementia risk factors, but how strongly are they linked? What predictive ability do we have in these domains? and how much variance in age and cognitive ability can they explain? These questions are often overlooked or ignored in the extant literature. Instead, neuroscientific questions are often prioritized, wherein researchers search for an age-related “lesion” by searching for voxels in the brain where gray matter variation statistically relates to an aging phenotype instead of asking how much variance in the aging phenotype is explained by brain measures (Reddan et al., 2017). This “lesion” model of imaging research in aging has led to useful insights, some reviewed above. However, now that substantial evidence exists that dementia risk often has widespread, global impacts on the brain (Douaud et al., 2014; Potvin et al., 2017), research is needed that integrates the widespread information available in imaging measures to predict dementia risk and develop interventions to prevent cognitive decline and dementia. This approach holds promise to help maximize our chances of developing sensitive, reliable, and modifiable biomarkers, where effect size and predictive ability can be prioritized.

One reason why formal prediction-studies using MRI are rare is because of intrinsic limitations in MRI data collection and measurement. In a typical structural MRI study, it is common to run over 100,000 different statistical tests (e.g., one for each gray matter voxel, which is typically 1 cubic millimeter) in the brain. But, due to the relatively high cost of MRI, it is quite rare for a study to even have 1,000 subjects, much less the hundreds of thousands required to rival the number of measurements in each subject. This imbalance, if improperly handled, can lead to overfitting and biased prediction estimates (Ioannidis, 2008; Kriegeskorte et al., 2010). To overcome these issues researchers are forced to either reduce the number of predictors (i.e., MRI measurements) or to estimate the predictive ability of a model with cross-validation with left-out, independent samples (Haynes, 2015; Poldrack et al., 2019). In order to formally identify dementia risk and accelerated aging in the brain, both of these approaches are becoming more widely adopted. Below I highlight representative studies that model these promising approaches in aging neuroscience.

While it is hard to find large MRI studies that have collected quality brain aging measures longitudinally across the lifespan, nearly every imaging study collects age, which is the largest risk factor for cognitive decline and dementia (Kawas et al., 2000). Researchers have begun to take advantage of the widespread collection of age as well as the emerging culture of data-sharing in neuroimaging (Mennes et al., 2013; Poldrack and Gorgolewski, 2014; Poline et al., 2012) to build predictive models of age with machine learning algorithms trained on structural brain measures (Cole and Franke, 2017; Koutsouleris et al., 2014). Machine learning algorithms can exploit and combine information from the hundreds of thousands of voxels in MRI images to predict age in left out datasets. This wealth of data has allowed rapid progression in our ability to predict age from MRI, with the best algorithms now able to explain over 90% of the variability in age across the lifespan from a T1-weighted structural MRI image alone (Cole et al., 2017b). The importance of these algorithms to age-related disease and dementia comes from the difference between an individual’s chronological age and their “brain-age” predicted by the machine learning algorithm.

This difference between an individual’s brain age and their chronological age is often referred to as brain Predicted Age Difference (“brainPAD”) (Cole et al., 2018). Individuals whose brains appears “older” than their chronological age have positive brainPAD values whereas individuals with relatively “younger” brains have negative brainPAD values. Revisiting our three complementary features of a midlife biomarker for dementia, brainPAD exhibits emerging support. First, age is by design an intrinsic feature of brainPAD, evidenced by its strong predictive ability of chronological age. Second, in multiple studies brainPAD has been correlated with cognitive ability and impairment. Individuals with dementia (Franke and Gaser, 2012) and Alzheimer’s disease (Franke et al., 2010) have higher brainPAD values (Cole et al., 2018) and as do those with greater cognitive decline (Elliott et al., 2019a; Liem et al., 2017). In addition, brainPAD is higher in people with diagnoses known to impair cognitive ability and accelerate aging, including schizophrenia (Koutsouleris et al., 2014; Schnack et al., 2016) and Down syndrome (Cole et al., 2017a). Third, brainPAD is associated with dementia risk factors including obesity (Ronan et al., 2016), diabetes (Ronan et al., 2016) and accelerated biological aging (Elliott et al., 2019a). Critically, in contrast to the traditional voxel-wise gray matter association studies, brain-age readily supplies effect size estimates and has proven to be a moderate predictor of pathological aging (Cohen’s D = 1.06 in dementia and .45 in mild cognitive impairment (Schmidt et al., 2011)). In addition, brainPAD predictions are constantly being refined with better algorithms, more diverse samples and improved input features. While early signs of success are accruing our ability to measure age-related structural change is still crude and often spatially limited to large white-matter tracts and large gray-matter parcels. As MRI-technology advances, the ability of brainPAD and other MRI biomarkers to measure subtle age-related changes with increasing precision promises to advance in parallel.

4.3. Open Questions and the Future of MRI-Based Midlife Biomarkers

While there is converging evidence that the effects of aging, cognitive ability, and dementia risk factors are associated with gray and white matter decline, there are many gaps in our knowledge of the longitudinal relationship between midlife aging and dementia. A major limitation is the current lack of prospective longitudinal studies with multiple time-points of cognition and MRI measurement. While there are large longitudinal MRI studies of age-related changes in brain, there are few, if any, studies that have longitudinal measurement of brain, cognitive ability, and dementia risk factors in the same individuals spanning midlife to old age. Without longitudinal measurements of all these variables across the lifespan in the same individuals, it is impossible to answer many of the most important remaining questions that could boost the confidence needed to use an MRI measure as a midlife biomarker for dementia risk.

Many of these open questions relate to the temporal relationships between gray and white matter changes, dementia risk factors, and cognitive decline. Further investigation of the compounding, interacting nature of dementia risk factors across the lifespan needs to be prioritized in future research. For example, does obesity lead to accelerated gray matter decline only if hypertension is present? Are their sensitive periods in which risk factors have exaggerated effects on brain integrity and dementia risk in later life? There is evidence to suggest that hypertension is more predictive of dementia in midlife (indicating chronic hypertension) than hypertension in old age (Launer et al., 2000), is this true of other risk factors and are these effects moderated by gray and white matter decline? Can subtle changes in MRI biomarkers be detected before accelerated cognitive decline can be detected? Can these biomarkers predict an individual’s risk of future cognitive decline before it happens? Longitudinal designs with lifespan measurements of dementia risk factors, cognitive ability, and brain imaging can begin to answer these questions.

The ideal study to fill in these gaps would be a prospective birth cohort study with repeated sampling of cognitive functioning, dementia risk factors, and brain imaging across the entire lifespan. This would allow fine-grained mapping of the interrelations between cognitive change, brain change, dementia risk factors, and ultimately cognitive impairment and dementia onset. However, this study has never been done due to resource limitations, as well as the historical fact that MRI has only been widely used in research for around 30 years (Fratiglioni and Wang, 2007; Meng and D’Arcy, 2012). While a study of this scale would require at least 80 years to complete and is unlikely to be available in the short-term, large samples with neuroimaging and cognition measured across the lifespan are becoming much more common including large-scale studies like the Human Connectome Project Lifespan study (Bookheimer et al., 2019).

4.4. The Potential of MRI-Based Biomarkers for Research and Midlife Interventions

Given the centrality of brain aging to cognitive decline and dementia, the continued development and refinement of MRI based biomarkers has unique potential to link midlife brain aging to future risk. More research is clearly needed because we are still far away from a set of sufficiently predictive, clinically relevant midlife biomarkers. However, synergistic refinement of midlife biomarkers aimed at improving their ability to predict accelerated aging and dementia risk while simultaneously using these biomarkers as outcome measures to push forward innovative midlife dementia prevention trials is one clear path forward. If this potential is realized, MRI-based biomarkers could identify individuals most in need of intervention and target them in midlife with aggressive prevention strategies before severe cognitive decline has set in. These interventions could then be closely studied and improved to minimize the rate of biomarker decline. Despite the challenges and large amount of research needed to realize this agenda, interventions to slow aging in midlife are likely to be one of the most cost-effective approaches to preventing dementia. If research into midlife geroprotection and MRI-based biomarkers are prioritized and continue to advance, dementia healthcare could adopt a focus on midlife prevention in addition to late-life management.

Figure 1.

Visual representations of the MRI biomarkers reviewed in the manuscript. A) cortical thickness is the distance between the gray matter boundary (red line) and the white matter boundary (yellow line). B) A visualization of the cortical surface, surface area is the size of each colored patch on the surface. C) A visualization of DTI white matter tracts. FA and MD are estimate by the same tensor model that estimates these tracts. D) A representative FLAIR image that is used to measure white matter hyperintensities. White matter hyperintensities can be seen as bring patches of white matter surrounding the lateral ventricles.