Risk for Alzheimer’s disease: A review of long-term episodic memory encoding and retrieval fMRI studies

Highlights

• Do separate risk factors for Alzheimer’s disease (AD) affect similar brain regions?

• We reviewed task-evoked fMRI studies of episodic memory in AD risk groups.

• AD risks were associated with non-localized, widespread brain activity alterations.

• The AD risk patterns were characterized by both greater and lower brain activity.

• Mixed fMRI findings do not indicate clear use as diagnostic tool.

Abstract

Many risk factors have been identified that predict future progression to Alzheimer’s disease (AD). However, clear links have yet to be made between these risk factors and how they affect brain functioning in early stages of AD. We conducted a narrative review and a quantitative analysis to better understand the relationship between nine categories of AD risk (i.e., brain pathology, genetics/family history, vascular health, head trauma, cognitive decline, engagement in daily life, late-life depression, sex/gender, and ethnoracial group) and task-evoked fMRI activity during episodic memory in cognitively-normal older adults. Our narrative review revealed widespread regional alterations of both greater and lower brain activity with AD risk. Nevertheless, our quantitative analysis revealed that a subset of studies converged on two patterns: AD risk was associated with (1) greater brain activity in frontal and parietal regions, but (2) reduced brain activity in hippocampal and occipital regions. The brain regions affected depended on the assessed memory stage (encoding or retrieval). Although the results clearly indicate that AD risks impact brain activity, we caution against using fMRI as a diagnostic tool for AD at the current time because the above consistencies were present among much variability, even among the same risk factor.

Introduction

Alzheimer’s disease (AD) is one of the major health problems in the United States, affecting nearly 6 million Americans above the age of 60 (Hebert et al., 2013), and its impact on the nation’s health care system will increase as the proportion of elderly in the population continues to rise, making prevention or delay of the disease onset crucial. Both neuroscientific and epidemiological approaches have identified proximal (in vivo pathology) and distal (e.g., biological, psychological, and health) risks for developing AD. What is less understood, however, are the factors that link these risks to alterations in brain function and neurodegeneration that ultimately lead to the cognitive and neuropsychiatric symptoms characterized by the disease. Functional magnetic resonance imaging (fMRI) might provide a link between distal and proximal risk factors that could not only provide insights to understand the mechanisms of the disease process, but also aid in early detection of preclinical AD in older adults. Here, we conducted a narrative review followed by a quantitative analysis to better understand the relationship between risk factors of AD and task-evoked fMRI activity. In our review, we focused on studies investigating episodic memory because these types of cognitive deficits have been most consistently found to decline longitudinally early in preclinical stages of the disease (Boraxbekk et al., 2015a; Grober et al., 2008; Mistridis et al., 2015; Schmid et al., 2013), are reliably associated with AD pathology (Hedden et al., 2013), and have been identified as the gold standard for characterizing AD (McKhann et al., 2011; Sperling et al., 2011).

The hallmark pathologies that characterize AD include tau neurofibrillary tangles and beta-amyloid deposition. Tau aggregation first begins to occur in the medial temporal lobes (MTL), whereas beta-amyloid plaques aggregate into clusters largely in the neocortex, especially in the medial prefrontal cortex and posterior cingulate (Braak and Braak, 1991; Nelson et al., 2012). These two forms of pathology appear to affect one another to become toxic and lead to synaptic and neuronal injury (for a recent review, see Aisen et al., 2017). Mice models of AD suggest that toxic beta-amyloid can lead to memory impairment (Leinenga and Götz, 2015; Morley et al., 2000). In contrast, studies in humans have found only weak associations between beta-amyloid and declines in cognition (Hedden et al., 2013), with many studies finding no associations (Amariglio et al., 2012; Doherty et al., 2015; Johnson et al., 2014; Perrotin et al., 2012; Song et al., 2016), suggesting a likely indirect link in humans. Instead, both mice and human studies have found strong associations between tau and cognition (Brier et al., 2016; Giannakopoulos et al., 2003; Santacruz et al., 2005). These associations are believed to be mediated by neurodegeneration as measured by regional cortical thickness or volume (Desikan et al., 2011; Dickerson and Wolk, 2013; Marks et al., 2017; Villeneuve et al., 2014).

Hypothetical models of the AD cascade have proposed that current neuroimaging technologies can detect the sequence of brain abnormalities associated with the progression of AD (Jack et al., 2013). In one such model, evident biomarkers proceed in the following order: (1) beta-amyloid, (2) tau, and (3) structural MRI/FDG-PET (Jack et al., 2013). Recent evidence has made further distinctions that hypometabolism (as measured by FDG-PET) appears to occur prior to brain atrophy (for review, see Habib et al., 2017). Where might fMRI play a role in this sequence?

On the one hand, fMRI may detect abnormalities in brain activity around the same time as structural MRI and FDG-PET. To the extent that AD pathology leads to synaptic injury (Sperling et al., 2010), brain activation as measured by fMRI should be sensitive to synaptic dysfunction (Logothetis et al., 2001). Researchers have argued that changes in brain activity using fMRI may occur before changes in brain structure are detectable (Mondadori et al., 2006; Mosconi et al., 2007), thus having a slight advantage for early detection compared with structural MRI. fMRI can also be used to help identify the regional pathways that are affected by elevated levels of pathology. For example, a debate exists as to which brain regions might be impacted by pathology first, such as the entorhinal cortex (Huijbers et al., 2014) or regions within the default mode network (DMN) (Buckner et al., 2009; Sperling et al., 2009). Differentiating between regions of early brain dysfunction not only informs the mechanisms and spread of brain abnormalities in the AD process, but can also inform which brain regions to target via tailored lifestyle interventions or pharmaceutical treatments (McDonough and Allen, 2019).

On the other hand, evidence is accumulating that synaptic abnormalities may actually precede the build-up of beta-amyloid (Jagust and Mormino, 2011; Jones et al., 2016), potentially placing techniques like fMRI toward the forefront of early detection measures. This new possibility builds on research showing that neural activity regulates the production of beta-amyloid in animal models (Kamenetz et al., 2003; Nitsch et al., 1993). This idea would imply that those brain regions with elevated levels of baseline neural activity or metabolic usage might produce more beta-amyloid in humans (Buckner et al., 2009). For example, young adults with genetic risk for AD (i.e., those who have the APOE ε4 allele) already show evidence of abnormal brain activity relative to young adults without genetic risk (Dennis et al., 2010; Filippini et al., 2009; Mondadori et al., 2007)—an age at which accumulation of AD pathology likely has not yet occurred. Although the findings are mixed as to whether more or less brain activity is apparent in preclinical stages of AD, the brain regions consistently implicated include the MTL and/or the DMN (Habib et al., 2017; Twamley et al., 2006). Researchers have argued that these brain regions may be most vulnerable to synaptic dysfunction because they are critical brain hubs and may most easily become “overworked” (Buckner et al., 2009; Jones et al., 2016; McDonough and Allen, 2019). However, regardless of whether brain activity precedes or follows AD pathology, fMRI could potentially be used as a tool to inform the mechanisms and staging of the AD process, including early detection.

Research has already shown wide-spread alterations in brain activity using fMRI in older adults with AD and MCI relative to cognitively-normal older adults. These studies have been summarized in several meta-analyses that have focused on fMRI activation during episodic memory tasks (Schwindt and Black, 2009; Terry et al., 2015; Wang et al., 2016). Despite the fact that multiple meta-analyses have been conducted that have used similar activation likelihood estimation methods and included many overlapping studies, few clear patterns could be gleaned across the meta-analyses. When compared to cognitively normal older adults, people with AD consistently exhibited lower brain activity in the MTL during both encoding and retrieval, whereas they exhibited greater brain activity in the lateral temporal cortex during encoding. Other regions of the brain including the prefrontal cortex, parietal cortex, and occipital cortex showed a mixture of patterns including greater, less, or no differences. The consistent lower activation of the MTL in those with AD has been interpreted as a primary cause of the severe episodic memory impairments, especially for recollective or contextual details associated with the disease (for reviews on recollection in AD, see Koen and Yonelinas, 2014; Schoemaker et al., 2014).

When comparing individuals with MCI to cognitively normal older adults, no clear patterns emerged across the meta-analyses: both greater and lower levels of brain activity were found in the prefrontal cortex, the MTL, the parietal cortex, and the occipital cortex. Differing patterns within the MTL have attempted to be reconciled by proposing that early or mild MCI is characterized by increases in brain activity (i.e., hyperactivity) relative to the preclinical AD stages, which then declines in later or moderate MCI, and precipitously declines thereafter (i.e., hypoactivity) when diagnosed with AD (Edelman et al., 2017; Ewers et al., 2011; Koelewijn et al., 2019; Merlo et al., 2019; Pasquini et al., 2019; Terry et al., 2015). Such early increases in activation in the MTL have consistently been suggested to be due to a type of compensatory processing. Specifically, the fMRI increases are thought to represent additional neural resources that are needed to compensate for neuronal loss (Dickerson et al., 2004; Grady et al., 2003). Similar compensatory explanations have been proposed to explain greater activity in the prefrontal cortex that sometimes accompanies lower MTL activity in both normal aging (Davis et al., 2008; Gutchess et al., 2005) and in the context of the AD process (Browndyke et al., 2013; Pariente et al., 2005; Wang et al., 2016). However, an alternative explanation for these increases is that it represents dysfunction or inefficiency of the neural circuitry (Logan et al., 2002; McDonough et al., 2013) and might be a sign of later cell death (Busche et al., 2008; Palop et al., 2007; Pasquini et al., 2019; Stern et al., 2004). Regardless of the exact mechanisms of these increases in brain activity, one might expect similar early increases in MTL activation in older adults at risk for AD, perhaps accompanied by increases in prefrontal cortex as well.

Numerous scientific fields have investigated risk factors for AD and dementia more generally, for decades. For example, the medical community has known that having a first degree relative (Huff et al., 1988; Mohs et al., 1987) and the apolipoprotein (APOE) ε4 allele (Corder et al., 1993; Saunders et al., 1993) confer risk for AD. Other non-modifiable risks include sex, such that women have a higher incidence rate of AD that men (for a recent review, see Buckley et al., 2018), and ethnoracial group, such that African Americans and Hispanics have a higher incidence rate of AD than non-Hispanic Whites (Perkins et al., 1997; Tang et al., 2001).

Epidemiological research also has shown that midlife and late-life health later predicts conversion to dementia, including AD (Brenner et al., 1993; Kivipelto et al., 2001). For example, a longitudinal population study from Sweden found that elevated blood pressure in one’s 70′s increased the likelihood of developing ADCE 15 years later (Skoog et al., 1996). These findings have been replicated in other populations, extended to risks found in midlife, and extended to other vascular health risks including high cholesterol (Kivipelto et al., 2001). Recent reviews have confirmed many of these well-known risk factors and have revealed more obscure factors such as diabetes, obesity, depression, smoking, heart disease, diet, head trauma, and renal dysfunction (Barnes et al., 2018; Deckers et al., 2015; Fann et al., 2018; Livingston et al., 2017).

Moreover, cognitive and social aging studies have shown that people who lack various types of lifestyle engagement have a heightened risk for developing AD (Deckers et al., 2015; Fratiglioni et al., 2004). For instance, people with less frequent participation in cognitive activity (Wilson et al., 2007), less social activity (Bassuk et al., 1999; Wang et al., 2002), less leisure activity (Crowe et al., 2003; Scarmeas et al., 2001), less physical activity (Angevaren et al., 2010; Buchman et al., 2012), or who experience a smaller degree of work complexity (Andel et al., 2005), have been observed to have a higher risk of developing AD. Relatedly, low levels of education (Mortimer and Graves, 1993) and low levels of cognitive reserve (Scarmeas and Stern, 2003) have also been associated with heightened AD risk.

As is evident in the above summary and the recent AD risk factor reviews, many risk factors are modifiable and give hope for the possible reduction of the AD incidence rate. However, only in the past decade have large-scale lifestyle interventions been implemented to help reduce the onset of dementia (Kivipelto et al., 2018), despite this information being available to intervention researchers and clinicians for more than 20 years. Hence, more work needs to be done to examine possible methods to delay AD incidence, onset, and progression.

Although prior research has shed light onto the risk factors that predict progression to AD and brain activation changes in AD, little is known regarding the link between risk factors and the early alterations in brain function associated with or predictive of the preclinical stages of AD. The present study aims to address this gap by explicitly examining the relationship between multiple risk factors for AD and brain functioning. Based on mostly epidemiological studies, we chose nine categories of risk factors that might begin to alter brain function in cognitively normal older adults. We then reviewed the literature for studies that assessed these risk factors using task-evoked fMRI of memory encoding and retrieval. From these studies, we also implemented a quantitative evaluation of the convergence of brain regions that might exhibit altered brain activity across these risk factors. According to a recent model of neurocognitive disorders (McDonough and Allen, 2019), these risk factors might lead to an upregulation of neural activity in hub regions of the brain, leading to a level of energy consumption that is no longer sustainable (for related ideas, see Buckner et al., 2009; Jones et al., 2016). These “overworked” brain hubs might then be more vulnerable to the buildup of AD pathology (e.g., amyloid and tau), thereby compromising neuronal processes that help restore and maintain normal functioning. Although each risk factor likely operates differently, this model proposes that the accumulation of these factors will converge on similar macro-level structures (i.e., brain networks) and lead to a systemic burden on critical brain hubs. Consistent with these ideas, we hypothesized that more brain activity would be found in the MTL and DMN regions for individuals with more AD risk, regardless of the category of risk (McDonough and Allen, 2019). Alternatively, these different risk factors might not converge on specific brain regions at all. Rather, multiple and different brain regions might be affected by each risk category, leading to brain-wide alterations of function. Although not mutually exclusive, the present study aims to provide evidence for either of these perspectives.