Abstract
Alzheimer’s disease (AD) and vascular dementia (VaD) are the two most common types of dementia. Although the combination of these disorders, called ‘mixed’ dementia, is recognized, the prevailing clinical and research perspective continues to consider AD and VaD as independent disorders. A review of recent neuropathological and neuropsychological literature reveals that these two disorders frequently co-occur and so-called ‘pure’ AD or VaD is comparatively rare. In addition, recent research shows that vascular dysfunction not only potentiates AD pathology, but that pathological changes in AD may subsequently induce vascular disorders. On the basis of these data, we propose that the neurobiological underpinnings underlying AD/VaD dementia and their neuropsychological phenotypes are best understood as existing along a clinical/pathological continuum or spectrum. We further propose that in conjunction with current diagnostic criteria, statistical modeling techniques using neuropsychological test performance should be leveraged to construct a system to classify AD/VaD spectrum dementia in order to test hypotheses regarding how mechanisms related to AD and VaD pathology interact and influence each other.
INTRODUCTION
Alzheimer’s disease (AD) and vascular dementia (VaD) are traditionally viewed as independent forms of dementia; and the amyloid cascade hypothesis [1, 2] continues to be the biological model explaining the mechanistic sequence of events that precipitate and ultimately lead to a clinical/pathological diagnosis of AD [3, 4]. According to this model, abnormal aggregation of β-amyloid (Aβ) in the brain initiates a cascade of events resulting in observed pathological alterations and subsequent neurodegeneration. Consequently, anti-Aβ therapies have been the predominant agent in AD clinical trials [5]. However, there are several important problems with the amyloid cascade hypothesis as the causative model for clinically diagnosed AD [6–8]. First, many individuals with a high amyloid burden do not exhibit clinical signs of dementia [9, 10]. Second, clinical trials using anti-Aβ therapies have significantly reduced Aβ burden from AD brain but failed to help or alter the course of AD in the treated patient [11, 12].
The failure of clinical trials to provide meaningful disease modification does not suggest that amyloid and/or other traditional types of AD pathologies are not important contributing factors to the clinical diagnosis of AD. However, there is now considerable evidence showing that both clinical and biological signs of vascular disease can potentiate the development of AD pathology [8]; and many patients diagnosed with either AD or VaD present with pathological and clinical heterogeneity (i.e., neurological, neuropsychological, brain imaging, and laboratory test evidence of both dementia syndromes) [13–16].
The failure of clinical trials to treat AD has been attributable to patient-selection strategies [5]. Responding to this critique, neuropsychologically-driven statistical modeling techniques applied to data collected as part of the donepezil/vitamin E clinical trial for mild cognitive impairment (MCI) showed better outcome [17]. Another suggestion put forth was to intervene with anti-Aβ agents earlier on in the disease process [5], a task that can be successfully accomplished also using statistically-driven neuropsychological methods [18, 19].
The point of view put forth in this paper is that based on well-established clinical heterogeneity associated with AD, operationally defined neuropsychological phenotypes should be leveraged and routinely used for classification in addition to the traditional means by which dementia syndromes are diagnosed. In other words, patients should be classified regarding how much or what proportion of a patient’s clinical or phenotypic presentation is associated with what type of underlying neuropathology by combining neuropsychological, imaging, and laboratory data [20, 21]. This type of algorithm for classifying AD/VaD spectrum dementia is very different from current dementia diagnostic schemes where the presence of a single biomarker or a positive finding from, say, a positron emission tomography (PET) imaging study defines a diagnosis [22].
We propose that classifying dementia syndromes and their prodromes can be achieved using statistical modeling techniques. The efficacy of this type of algorithm is currently gaining traction in the literature [18, 23–25] and is particularly relevant for selecting patients and formulating expected outcomes for clinical trials designed to treat dementia [17]. In the sections below, we describe the extent of clinical heterogeneity in the disease process through a review of how vascular disease has been shown to potentiate traditional AD pathology and evidence for neuropathological and neuropsychological heterogeneity. Finally, we will present evidence in support of classifying patients using neuropsychological statistical modeling techniques and how neuropsychologically-based approaches can elucidate the brain-behavior relations.
SYNERGISTIC RELATIONS BETWEEN VASCULAR DISEASE AND ALZHEIMER’S DEMENTIA
Cardiovascular risk factors and dementia
It is now well known that common cardiovascular risk factors are highly correlated with the eventual emergence of AD [26] including midlife hypertension [27, 28], hypercholesterolemia [29, 30], diabetes mellitus [29, 31], midlife obesity [32], compromised cardiac functioning and cerebral arterial disease [33, 34], smoking [35], and physical inactivity [35]. Clark and colleagues [36] studied a group of community volunteers with abnormal Aβ levels via PET or cerebrospinal fluid (CSF) with and without vascular risks such as hypertension and obesity. Patients with both vascular risk and abnormal Aβ performed worse on immediate and delayed free recall on a serial list learning test such that the presence of vascular risk factors moderated poorer performance.
Impaired insulin signaling, hyperlipidemia, the APOE4 allele, reduced glucose availability, and cerebral blood flow have all been implicated as potential initiators and facilitators of AD-related neurodegenerative dysfunction [37–41]. Diabetes has been linked to reduced insulin transport across the blood-brain barrier (BBB), insulin resistance, and impaired insulin signaling [31, 42]. In a normally functioning brain, insulin is readily transported across the BBB through a receptor-mediated process, and influences vasoreactivity, capillary recruitment, vasodilation, and regional blood flow [15, 43–45]. Endothelial dysfunction due to insulin resistance reduces insulin transport, capillary recruitment, and microvascular blood flow, which then interferes with neurovascular unit functioning and the coordinated interactions of astrocytes, neurons, and endothelium that augment vasoconstriction and inflammation [29, 30].
Hyperlipidemia is also known to contribute to small vessel disease as well as a clinical diagnosis of AD [32, 46]. Lipids and lipoproteins play a critical role in Aβ production and clearance, and dyslipidemia is associated with insulin resistant syndrome, as insulin is a primary regulator of lipid metabolism. Results from a diabetic-hypercholesterolemic (DMHC) porcine model suggest that DMHC increases BBB permeability and influx of plasma components into the brain tissue, thus disrupting brain homeostasis and propagating intracellular Aβ deposition [37].
APOE functions as a lipid transport protein ligand for low-density lipoprotein (LDL receptors), and plays a role in cholesterol metabolism [47, 48]. APOE4 is associated with increased BBB damage (e.g., astrocytes and tight junctions of vascular endothelial cells and cerebrovascular amyloid angiopathy), and negatively affect clearance of Aβ from the brain via upregulation of receptor for advanced glycation end-products and downregulation of LDL receptor-related protein 1 [41, 50]. APOE4 carriers also have increased pro-inflammatory levels that degrade BBB tight junctions and basement membrane that cause secondary neuronal dysfunction [41]. Finally, hypertension has been shown to impair functional hyperemia, i.e., the way in which brain activity and blood flow are synchronized [29, 32]. Hypertension impacts neuronal function by increasing microvascular pressure, which increases BBB disruption and plasma influx into the brain, ultimately leading to impaired clearance of protein [51].
Alterations in blood-brain barrier
BBB breach can be the sequela of chronic cardiovascular disease. The neurovascular unit is composed of endothelial cells, pericytes, vascular smooth muscle cells, glia, and neurons, which together act to control BBB permeability and regulate blood flow, and are vital components needed for the proper maintenance of brain homeostasis and the neuronal milieu. Homeostatic dysregulation of elements related to the neurovascular unit can disrupt protein-clearing systems in the brain resulting in disturbed neuronal functioning. Chronic worsening of this situation can eventually overwhelm any compensatory mechanisms and trigger neuropathological alterations resulting from the accumulation of toxic molecules due to lack of proper clearance. It is, therefore, not a coincidence that the areas of the brain where AD pathology is seen (i.e., hippocampus, temporoparietal cortex, and anterior cingulate cortex) are the areas of the brain where the BBB is most commonly disrupted.
There is also research suggesting that vascular dysregulation including BBB dysfunction may be among the earliest markers of pathology that ultimately lead to a diagnosis of AD. For example, Iturria-Medina and colleagues [52] sought to characterize the temporal ordering of factors related to the eventual emergence of AD, including Aβ deposition, glucose metabolism, cerebral blood flow, resting functional magnetic resonance imaging (MRI) neuronal activity, tissue properties (structural MRI), neuropsychological test performance, and peripheral protein levels (CSF & plasma). These investigators found that vascular dysregulation within the brain was the earliest marker for pathological events, followed by poor performance on a memory test leading to AD suggesting that BBB dysfunction may be an important factor propagating neurodegeneration.
To help clarify the relations between neurocognitive decline in older adults and neurovascular dysfunction, Nation and colleagues [53] studied brain capillary damage using a novel CSF biomarker of BBB (i.e., capillary mural cell pericyte, soluble platelet-derived growth factor receptor-β8), and regional BBB permeability defined with dynamic contrast-enhanced MRI. Using objective neuropsychological metrics of cognitive impairment derived by Jak et al. [54], these researchers found that older adults with neurocognitive dysfunction had BBB breakdown with pericyte injury in the hippocampus regardless of tau or Aβ changes, suggesting that BBB breakdown is an independent biomarker and not necessarily related to dysregulated tau and Aβ. The two studies described above are examples of how neuropsychologically-based measures can be used to provide sensitive or reliable representations of the AD/VaD disease process.
NEUROPATHOLOGY HETEROGENEITY IN AD/VaD SPECTRUM DEMENTIA
There is a growing body of research demonstrating considerable heterogeneity with respect to the neuropathology that is present in AD/VaD spectrum dementia [55, 56]. For example, recent evidence suggests that, upon autopsy, approximately one third of patients diagnosed with VaD showed evidence of basal forebrain cholinergic deficits, a problem traditionally linked with AD [57]. Impaired cholinergic transmission in VaD include a reduced number of cholinergic nerve terminals, decreased activity of choline acetyltransferase (enzyme for acetylcholine synthesis) in the cerebral cortex, hippocampus, and striatum, as well as degeneration of nicotinic cholinergic neurotransmission and receptors [57].
Recent epidemiological and community-based studies clearly show mixed neuropathology in many, if not the majority, of the patients diagnosed with dementia [20, 58–65]. For example, Wharton et al. [66] and Schneider et al. [64] found mixed AD/VaD pathology to be quite common, if not the norm, in their community-based samples. Boyle and colleagues [59] examined the brain pathology associated with cognitive loss in over 1,000 elderly individuals, including AD pathology (i.e., neuritic plaques, diffuse plaques, and neurofibrillary tangles), neocortical Lewy bodies, hippocampal sclerosis, TDP-43 cytoplasmic alterations; and vascular lesions including gross infarcts, microinfarcts, moderate-severe arteriolar sclerosis, moderate-severe atherosclerosis around the Circle of Willis, and moderate-severe cerebrovascular amyloid angiopathy. They reported that AD pathology without any other neuropathology was seen in only 9% of their sample, thus, rarely seen in isolation. More often, AD pathology was seen in combination with either vascular pathology and/or other indications of neurodegeneration. Other studies have found similar results [67, 68]. In fact, Silbert and colleagues [68] noted that of the 63 patients who converted to MCI and underwent autopsy, only 28% had AD as the single cause of their dementia. A more or less equal number of patients (24%) presented with both AD and significant vascular alterations.
These results have been echoed by Kling et al. [20] who have suggested the need for an integrative rather than a taxonomic approach to the interaction of vascular and AD-related pathology. These researchers suggest that, instead of attempting to segregate AD versus VaD, there is a need to comprehensively understand the underlying pathophysiology of various dementia phenotypes and their concomitant neuropathologic characteristics. In view of the fact that neuropathological overlap is clearly present, there is an immediate need for research that clarifies how neurodegenerative and vascular pathologies interact and potentiate each other.
NEUROPSYCHOLOGICAL PHENOTYPES IN ALZHEIMER’S/VASCULAR SPECTRUM DEMENTIA
Statistical modeling of AD/VaD spectrum neuropsychological phenotypes
There are many advantages to using statistical techniques to classify AD/VaD spectrum patients into their meaningful groups including 1) the ability to use statistically-determined phenotypes as a grouping or independent variable; 2) the objectivity in operationally-defining underlying neurocognitive constructs associated with dementia groups; and 3) the capacity for replication permitting comparisons between studies. Three studies undertaken by Libon and colleague are emblematic.
Price et al. [69] and Libon et al. [70] used a visual rating scale [71] to to re-classify patients diagnosed clinically with either AD or VaD into three groups—patients with mild, moderate, and severe MRI white matter abnormalities (WMAs)—and constructed neuropsychological indices measuring executive control/visuoconstruction versus episodic memory/language abilities. Four major findings emerged from this research. First, there was a double dissociation between patients re-classified as presenting with mild versus severe MRI WMAs. Patients with mild WMAs clearly performed worse on the episodic memory/language index, while patients with severe WMAs were particularly impaired on the executive control/visuoconstruction index. Thus, based on severity of MRI-determined WMAs, there appeared to be two distinct single-domain neuropsychological syndromes. Second, patients who presented with moderate WMA were equally impaired on executive control/visuoconstruction and episodic memory/language indices. Thus, these patients can be viewed as presenting with a multi-domain neuropsychological syndrome. Third, of the three MRI-determined WMA groups, the multi-domain group contained the largest number of patients, suggesting that clinical heterogeneity is quite common. Fourth, when each neuropsychological index was analyzed within group, the multi-domain patients tended to produce scores that were less impaired than their single domain counterparts.
These data were largely replicated by Libon et al. [72] where latent class analysis found that AD/VaD patients could be classified into four distinct groups: two single domain groups typified by executive impairment and amnesia, respectively, and two multi-domain groups where neuropsychological impairment tended to be equally distributed across the neuropsychological domains analyzed. Subsequent analysis of errors [73, 74] found that the single domain amnestic patients presented with classic features amnesia but were less impaired on other neuropsychological tests. By contrast, the single domain dysexecutive group was remarkable for a retrieval or source-recall problem on memory tests [75] without striking evidence for actual amnesia. In addition, there was evidence for a pervasive dysexecutive syndrome across all neurocognitive domains assessed. Moreover, there was a relative dissociation such that the latent class analysis-determined dysexecutive group presented with greater MRI white matter alterations single domain amnestic patients presented with smaller hippocampal volume.
Data from these experiments suggest that statistically-defined neuropsychological phenotypes can inform brain regions that are compromised and the distribution of underlying neuropathology. It is possible that using the statistical techniques described above to recruit patients for clinical trials designed to treat dementia, may result in better treatment outcomes.
STATISTICAL MODELING OF NEUROPSYCHOLOGICAL PHENOTYPES, DEMENTIA BIOMARKERS, AND IMPLICATIONS FOR TREATMENT
There is new research providing criterion and external validity linking classification of patients with suspected dementia using the statistical-modeling techniques with biomarkers for dementia. Bondi and colleagues have conducted a series of studies on patients with MCI, a prodrome that often leads to dementia such as AD. These researchers have provided considerable evidence showing the utility of statistically-determined neuropsychological groups as a way of linking MCI phenotypes with dementia biomarkers, including reduced hippocampal volume in amnestic versus non-amnestic MCI groups [76], fluid biomarkers related to AD (e.g., Aβ1-42, total tau, and p-tau181/Aβ1-42; [77]), and patterns of MRI cortical thinning [78]. These neuropsychological methods may be helpful in uncovering divergent and convergent neurodegenerative processes in the hopes of making progress toward person-centered treatments.
In research with considerable clinical implications, Edmonds et al. [17] leveraged neuropsychological test performance using statistical modeling techniques in their re-analysis of the original donepezil/vitamin E trial designed to treat MCI [79]. The original trial defined MCI using the Petersen criteria [80, 81] and reported only modest treatment effects. Edmonds et al. [17] used cluster analysis to re-classify participants into groups presenting with amnestic MCI, mixed MCI, and normal cognition. Subsequent analyses of the effect of drug on neurocognition after removing the statistically-determined cognitively normal group revealed substantially lower rates of progression to dementia and better performance on memory tests. Thus, Edmonds et al. [17] found a better treatment outcome as compared to the original research findings.
These data, along with the neuropsychological investigations described above, underscore the efficacy of statistical modeling techniques to characterize neuropsychological test performance and have important implications for all clinical trials designed to treat dementia. The failure of recent clinical trials to treat AD may be due to differing and/or unreliable classification of individuals into dementia and prodromal groups [17]. An alternative procedure is to use statistical modeling techniques that operationally define highly nuanced AD/VaD spectrum phenotypes. This type of classification may provide greater information regarding underlying biomarkers associated with neurodegeneration.
COMBINING MRI AND NEUROPSYCHOLOGY TEST PERFORMANCE
A means by which to classify AD/VaD spectrum dementia and MCI patients using a statistical modeling technique has been suggested by Brickman et al. [6]. In this study, statistical techniques were used to construct an MRI-based quantitative index measuring both vascular alterations (i.e., white matter hyperintensities and infarcts) and AD-type changes (i.e., hippocampal volume and cortical thickness) associated with performance on a verbal serial list learning test. These researchers found expected relations when healthy controls, MCI, and dementia patients were analyzed such that the MRI index was lower in AD compared to MCI patients. Moreover, the MRI index was correlated with PET amyloid imaging and CSF levels of total tau, phosphorylated tau, Aβ in life, and postmortem neurofibrillary tangles, neuronal loss, atrophy, and infarcts upon autopsy. Finally, this MRI index predicted conversion to MCI and AD from healthy controls. All of the MRI measures of interest contributed to the model. This type of methodology has potential to directly assess the contribution of MRI indices associated with dementia as related to neuropsychological test performance. Although only episodic memory test performance was used in this research, other domains of neuropsychological functioning such as executive control could also be used. Similarly, other MRI measures such as volume of important subcortical nuclei in addition to the hippocampus could be used. This research provides an operationally defined means that directly assesses variance accounted by neuropsychological test performance related to MRI markers of dementia. This approach could easily form the basis of systems for classifying patients where there is relatively greater or equal contribution of underlying neuropathology alterations as related to neuropsychological test performance. The methods suggested by Brickman et al. [6] are consistent with ideas proposed by Cosentino et al. [82]. Specifically, a more heuristically meaningful way to describe AD/VaD spectrum patients would be on the basis of separate indices examining, say, the degree of episodic memory versus executive impairment; and the relative contribution of gray versus vascular alterations as seen on MRI scans.
CONCLUSION
Results from clinical trials using anti-Aβ immunotherapies have been disappointing. The amyloid cascade hypothesis has failed to incorporate other important mechanisms that drive pathology, like the contribution of intraneuronal amyloid deposition seen in human and AD transgenic mouse brains into models of AD pathogenesis and plaque formation [83–85]. Reducing extracellular amyloid without addressing intraneuronal amyloid deposition is unlikely to improve neuropsychological performance. Of course, no amount of amyloid removal can restore function to cells that have already died and released their content of amyloid that contribute to the formation of plaques [86]. This is at least one reason why leveraging neuropsychological tools to flag patients earlier on in the disease process is all-the-more necessary.
The data that has gathered over past decades suggest that alterations induced by underlying vascular risk factors such as hypertension, diabetes, hyperlipidemia, and heart disease may not only be responsible for the structural lesions seen on CT or MRI scans and impairments in cognitive function; but also induce other changes, such alterations in the BBB, that may be the initial, upstream insult resulting in AD/VaD spectrum neurodegeneration. Thus, the biological mechanisms associated with AD and VaD, that were once thought to be independent, are intimately related. The use of spinal and serum fluid biomarkers as well as tau and amyloid brain imaging are valid to the extent that these technologies can measure the presence of proteinopathies traditionally linked to AD. However, vascular risk factors and imaging evidence of vascular injury to the brain must be afforded equal importance as the AD fluid/imaging biomarkers in order to fully understand and appreciate what can only be described as a highly contextualized, multi-dimensional view of AD/VaD spectrum dementia. The degree to which both vascular and traditional-AD markers are prevalent is supported by postmortem studies that reveal combined VaD and AD pathology to be very common and that ‘pure AD’ or ‘VaD’ syndromes are infrequent. Rather than viewing AD and VaD as the two most common types of dementia, it may be more accurate to say that dementia-related proteinopathies with vascular disorders are the most common mechanisms underlying insidious onset dementia. Yet, the ability to robustly find cognitive and behavioral sequela of the disease process is contingent on the way in which we define our groups.
In this paper, we have emphasized how statistical modeling techniques are able to differentiate highly nuanced single domain and multi-domain AD/VaD spectrum syndromes. A wealth of information regarding the brain-behavior relationships that underlie these syndromes is now available [87–89]. Leveraging statistically defined AD/VaD phenotypes to study BBB alterations, cardiovascular disease, and MRI alterations on the brain may help clarify and develop a more effective classification of AD/VaD spectrum dementia syndromes, perhaps leading to more effective means of treatment and prevention.
The traditional use of neuropsychological assessment in dementia research has primarily been to assess outcome. This needs to change. As demonstrated by Edmonds et al. [17], statistically-determined neuropsychological assessment can and should be used as an independent variable to classify patients for treatment. The neuropsychological research reviewed above clearly demonstrates how one can leverage theoretically meaningful constructs to define empirically meaningful single and multi-domain phenotypic syndromes. In sum, we maintain that classification of the neuropsychological syndromes using neuropsychological methods should be integrated into current AD/VaD diagnostic schemes. This kind of paradigm maximizes and puts to best use all available data to manage and treat what has become the leading public health problem of our age.
CONFLICT OF INTEREST
Authors’ disclosures available online (https://www.j-alz.com/manuscript-disclosures/19-0654r1).