Neuropathology and Cognitive Impairment in Alzheimer Disease: A Complex but Coherent Relationship

Specificity of AD Lesions: Amyloid Plaques and NFTs are Seen Outside of AD

It has been suggested that neurofibrillary pathology and amyloid plaques are not specific to AD. If this is true, does nonspecificity argue for or against the importance of NFTs and amyloid plaques in AD?

Neurofibrillary tangles are not specific for AD, particularly if a broader definition of NFTs includes different tau isoforms or if one expands the expectations of the morphological characteristics of NFTs (2, 3). In addition to AD, NFTs are also found in some frontotemporal dementias, myotonic dystrophy, viral panencephalitis, dementia pugilistica, some prion diseases, and other brain diseases (4, 5). For many of these disorders, the severity of NFT pathology is less than that observed in end-stage AD. No condition characterized by widespread neocortical NFTs lacks extensive neurodegeneration and clinical dementia. On the other hand, there are many subtypes of chronic brain diseases in which there are extensive neurodegeneration and clinical dementia without NFTs, such as many subtypes of frontotemporal dementias, synucleinopathies, subacute or chronic infarcts, metabolic, demyelinating, developmental, and trinucleotide repeats diseases. Nor is there any doubt that tau protein itself can directly trigger neurodegeneration: many germ line mutations in tau produce clinical dementia with NFTs (3, 6). These tauopathies are distinct from AD, but common pathways may be involved. The bottom line is that NFTs appear in multiple brain diseases, and some researchers, including the present authors, infer that NFTs contribute to neurodegeneration in more than 1 disease state.

Unlike NFTs, the appearance of amyloid plaques in AD brains is unique. Dystrophic neurites that contain amyloid precursor protein are seen in traumatic brain injury (7, 8), and “diffuse plaques” can be observed in association with dementia pugilistica (9). “Lewy plaques” have also been described, with a corona of dystrophic α-synuclein–positive neurites (10). By contrast, the particular appearance of neuritic plaques (i.e. Aβ peptide–containing extracellular lesions surrounded by tau neurofibrillary pathology) is considered to be specific for AD. Amyloid plaques are not a nonspecific reaction to neurofibrillary pathology because non-AD tauopathies lack amyloid plaques. Further attesting to the specificity of AD-type amyloid plaques is the fact that mutations or duplications in the amyloid precursor protein gene produce the specific features of AD, clinically and pathologically (11–13). Notably, without NFTs or amyloid angiopathy, amyloid plaques are not associated with neurodegeneration (see later). In sum, AD involves a specific combination of neuritic amyloid plaques and NFTs; the neuritic plaques are more specific to the disease, and the NFTs seem more likely to induce neurodegeneration.

Specificity of AD Lesions: General and Theoretical Considerations About the Clinical Context of AD, Neuroanatomy, and Synergy Between Plaques and NFTs

One challenge of CP studies in AD is to place nonlinear disease processes onto quasi-linear continua for purposes of correlation. The pathological changes in AD follow a complex stereotyped neuroanatomical pattern (14). Prior studies have produced a general scheme for how the pathology correlates with clinical outcomes (Fig. 2 and Fig 3), but it is important to consider the clinical context for this association.

Epidemiological studies indicate that clinical AD prevalence doubles every half decade after age 65 years so that approximately 25% of 90-year-olds carry the diagnosis of AD (15, 16). The time from formal clinical diagnosis to death is only approximately 8 years (17). Structural and cognitive disease is present over a longer time span, however, certainly over a decade before death and perhaps considerably longer (18–24). Using these data, we deduce that an average person dying at age 78 years will have approximately 5% to 10% likelihood of having an antemortem diagnosis of AD, but at least approximately 20% to 40% likelihood of having significant AD-type pathology. Thus, many individuals with AD-related pathology and minimal or no detectable antemortem cognitive deficits die of other causes. Theoretical and empirically derived data agree on this important point, as shown in Figure 4. Figure 4B shows data of Del Tredici and Braak (25) based on 3,928 cases. The existence of nondemented persons with AD-type pathology is hence not a potential confound; it is obvious that many persons die in the preclinical phase of AD. Preclinical and subclinical diseases are well-accepted ideas in cancer, atherosclerosis, and in many other diseases, but seem to cause confusion in discussions of AD.

In addition to the concept of preclinical AD, there are other potential pitfalls in CP studies. Distinct neuroanatomical regions are affected at different stages of AD, and this must be taken into account for CP correlations. For example, portions of the allocortex (Table 1), including the hippocampus and entorhinal cortex, may be involved severely early in AD (14, 26–28) when neocortical areas are still relatively well preserved. The involvement of neocortical areas presumably underlies the later-stage cognitive impairment. Therefore, study of hippocampus may be best for assessing earlier memory changes, but a study limited to hippocampus would be inappropriate for correlating overall pathology with overall cognitive decline. An important case study is that of Patient H.M., who had bilateral near-complete temporal lobectomies in 1953 (29, 30). Although mostly amnestic, Patient H.M. is still quite robust cognitively more than 5 decades later without AD (30) because most cognitive domains impaired in AD pertain to neocortical involvement (i.e. nondeclarative memory, judgment, gnosia, visuospatial skills, language, executive function, etc.).

As if factoring in both the clinical course and the anatomical aspects of the disease was not difficult enough, there is added complexity in the lesions of AD. By definition, AD is characterized by the presence of both amyloid plaques and neurofibrillary pathology, and there seems to be some synergy of the two, most definitely in neuritic plaques, but the tracking of 2 separate pathologies renders CP correlations challenging. In the multidecade course of AD progression, amyloid plaques and NFTs may be first observed in different parts of the brain: NFTs tend to appear first in allocortical structures, whereas amyloid plaques may first be found in the neocortex (26, 27, 31–35). Because 2 different types of pathology develop initially without apparent direct correlation to each other, a linear continuum is difficult to establish. A further layer of complexity is that amyloid plaques and NFTs are probably not static lesions (see later).

These considerations help explain why some prior studies have shown apparent general discrepancies with regard to the correlation of extent of pathology and the severity of premortem cognitive decline (1). However, apparent incongruities have been asserted by a number of researchers, and a discussion is provided later regarding specific studies.

Specificity of AD Lesions: Particular Cases and Pathological Combinations

AD is defined by the presence of neuritic plaques and neurofibrillary pathology, the abundance of which is hypothesized to correlate with the severity of cognitive deterioration. If this is so, then certain corollaries must hold true: 1) there should not be individuals with extremely abundant plaques and tangles but no cognitive deterioration; 2) there should not be clinical dementia cases with abundant plaques and no NFTs; 3) there should not be patients with clinical dementia in whose brains there is no detectable pathological substrate; and 4) although there exist a number of tauopathies (progressive supranuclear palsy, corticobasal degeneration, Guamanian parkinsonism, FTDP-17, etc.), there should not be a true “tau-only” AD. These ideas have been tested repeatedly and have fostered controversy. Different specific groups will be discussed in turn.

Amyloid Plaques and Neurofibrillary Pathology But No Dementia

Cases with considerable AD-type pathology but without documented cognitive decline have been termed pathological aging and presymptomatic aging (36–38). Because mild cognitive impairment is considered to constitute an early expression of AD (39), we here discuss only nondemented patients. Such patients with incipient AD-like changes have also been called preclinical AD based on magnetic resonance imaging data (19, 40), neuropsychological testing (21, 41), Pittsburgh Compound B positron emission tomography imaging (42), and cerebrospinal fluid biomarkers (43).

Pathologically, many nondemented cases meet the older Khachaturian diagnostic criteria (44) and a significant minority meet Consortium to Establish A Registry for Alzheimer’s Disease definite criteria for AD (41, 45). These criteria have been superseded for the stand-alone AD diagnosis since 1997 (46). Moreover, these criteria are not relevant directly to a discussion of false-positives because they were established for making the diagnosis of AD in clinically demented patients.

The pathological diagnosis of “intermediate likelihood” for AD by the current National Institute on Aging–Reagan Institute (NIARI) criteria corresponds to Braak Stages III and IV (46), neither of which is by itself a substrate for profound cognitive impairment (Fig. 2). The NIARI criteria for pathological diagnosis of “high likelihood” for AD correspond to Braak Stages V and VI (46). Note that the average final Mini-Mental State Examination (MMSE) score for Braak Stage V is approximately 20 out of a possible 30 (Fig. 2), so may include some relatively normal individuals. A recent case report described a 92-year-old nondemented patient whose brain, on autopsy, contained Braak Stage VI pathology (47), but neocortical NFTs from this patient did not appear particularly dense, and the patient did demonstrate cognitive deficits. In 3 different cognitive tests, she scored at or below the 20th percentile for her age despite previously normal test results (47). Because there is as yet no published report of true end-stage pathology in a cognitively intact individual, an assessment of larger studies is necessary.

In 2004, Fernando and Ince (48) described a population-based autopsy series in which “…severe neocortical NFTs were not seen in any nondemented individual.” This result has been obtained repeatedly from multiple centers (33, 47–57), as shown in Table 2 (N = 555 nondemented subjects, of whom 15 were Braak Stages V or VI). These data from many research centers indicate a 2.7% false-positive rate in the NIARI criteria for high-likelihood AD diagnosis, with mostly Braak Stage V cases. In addition, a number of other studies have noted that nondemented patients have NFTs confined predominantly to the mesial temporal lobe structures (58–61). These data support the accuracy of the high likelihood NIARI criteria, but leave unanswered the important question: are there nondemented persons with truly end-stage AD-type pathology?

We addressed this issue using data from the University of Kentucky Alzheimer’s Disease Center autopsy series (62, 63). Of 59 longitudinally followed nondemented subjects with MMSE scores within a year of death, 9 showed Braak Stage IV, 3 Stage V, and 1 Stage VI (Table 2). The single Braak Stage VI patient from this series had a final MMSE score of 29 within a year of death. Note that in this brain, there was a relatively low density of neocortical NFTs (Fig. 5, rightmost data points). This person seems to demonstrate the lowest-severity Braak Stage VI. Thus, even among Braak Stage VI cases, individuals with lower neocortical NFT densities tend to have better preservation of cognitive faculties. Variability in Braak Stage VI severity is underscored by different studies in which the median final MMSE scores range from less than 2 (64) to 14 (53) in this group. Moreover, in a study of 168 autopsied persons (Fig. 6), every case with neocortical NFT density greater than a particular threshold (blue arrow) had clinical dementia (MMSE, <20), just as every case with low or absent neocortical NFTs (red arrow) had final MMSE scores greater than 20. In sum, whereas the correlation between AD-type pathology and clinical impairment is not absolutely predictable in each patient, there seems to be a threshold of pathology above which every patient, without exception, is profoundly impaired.

“Tangle-Only” Dementia

There are reports of autopsied patients with neurofibrillary pathology–predominant dementia (5, 54, 65, 66). These cases are characterized by abundant NFTs in medial temporal lobe structures with no or few amyloid plaques throughout the brain. The individuals are usually older and primarily had memory decline and reduced activities of daily living. Whether this is a subtype of AD or a different entity remains to be elucidated.

Other diseases are now known to be characterized by NFT formation and minimal or absent amyloid pathology. These cases are now presumed to reflect tauopathies, which are a diverse group of diseases that may be associated with genetic (tau mutations) or environmental (Guamian or postencephalitic parkinsonism) factors (3, 5). In most population-based series, and in older patients, tauopathies are uncommon (54, 57, 67).

“Plaque-Only” Dementia

We have no personal experience of cases with dementia, abundant neocortical neuritic plaques, or diffuse plaques but with no other clinical or pathological explanation for the cognitive decline. The phenomenon of plaque-only dementia has been described (68, 69) but may partly correspond to patients with dementia with cortical Lewy bodies (70). Even in the recent series of plaque-only AD (68), which described a group of 16 brains of persons with clinical dementia but with no neocortical NFTs and a low number of amyloid plaques, it is unclear whether or how other prevalent brain diseases that are capable of contributing to dementia were excluded. Thus, with the apparent exception of a rare familial variant of historical significance (71), plaque-only dementia has not been demonstrated beyond doubt despite many studies addressing this topic. Without fuller clarification, it is possible that the small numbers of plaque-only dementia cases described to date had cognitive decline caused by other neurodegenerative conditions.

Cases With Dementia Lacking Amyloid Plaques, NFTs, and Other Pathological Substrates

The literature does not indicate an appreciable subset of patients with clinical dementia and histopathologically pristine brains on autopsy. Studies that show a significant proportion of cases with dementia, low neurofibrillary pathology, and no other known pathological substrate, tend to be the older studies that lack probes for recently described diseases. In 2000, Crystal et al (72) described a cohort with “dementia of unknown etiology,” however, cognitive impairment in this cohort was not extreme, and these patients showed specific contributory neuropathological features including hippocampal sclerosis and vascular disease.

There are many additional causes of cognitive decline, and these can be clinically diagnosed in most cases. For this reason, in clinical series such as those at the University of Kentucky Alzheimer’s Disease Center, which include rigorous longitudinal clinical assessments followed by brain autopsy, extremely rare patients with profound dementia and the premortem diagnosis of AD have lacked histopathologic substrates at autopsy. We and others have found that persons with concomitant pathologies die with less neurofibrillary pathology (63, 64). Identifying and taking into account the presence and magnitude of concomitant pathologies is probably the biggest challenge to studies of the aging human brain.

Concomitant Pathology: The Large Impact of Cerebrovascular Diseases

The most prevalent concomitant pathology in the brains of aged persons, cerebrovascular disease (CVD), is directly relevant to any discussion of CP studies in dementia or AD. The challenges introduced by this widespread but unpredictable comorbidity in aged persons have been discussed previously (48, 64, 73–79). Yet, the profound impact of CVD on studies pertinent to cognition in the elderly seems to be underappreciated, or even ignored, in the broader field of AD research.

Cerebrovascular disease spans a wide spectrum including small- and large-vessel ischemic disease, embolic, venous, inflammatory, hypertensive, hemorrhagic, aneurysmal, hypoglycemia, hypoxia, amyloidogenic, and others. The importance of CVD is at least 2-fold. First, some cerebrovascular abnormalities are so common as to be the norm in advanced old age. Second, CVD induces an unpredictable change in cognition and hence presents challenges to correlating the severity of cognitive decline with other diseases.

Clinical and subclinical CVD is common in elderly subjects. Cerebrovascular disease is detectable in 75% to 90% of persons older than 90 years (63, 80), and whereas frank clinical strokes affect approximately 750,000 in America each year, more than 11 million discrete but clinically silent cerebrovascular events are thought to occur over the same interval (81). The high frequency of cerebrovascular changes among elderly patients means that both cognition and other pathology need to be weighed in the context of an altered milieu. Furthermore, unlike AD, CVD progresses in an anatomically, clinically, and temporally unpredictable fashion, and there is no systematic methodology for correlating pathological and clinical findings (82). These are some reasons why CVD-related pathology is frequent in cases termed dementia of unknown origin (72).

Importantly for CP studies, CVD shifts the detection threshold for cognitive changes during life. In elderly AD patients, CVD leads to more rapid cognitive deterioration (83, 84). Subcortical and/or lacunar infarcts alter the detection threshold for dementia (74, 85–87). The change in clinical course is mirrored by a different pathological appearance in the brain. Persons with CVD tend to have lower AD-related pathology with a given degree of cognitive impairment (63, 86). In other words, the frequent presence of concomitant CVD severely dampens the association for other pathologies, such as neurofibrillary pathology and amyloid plaques, to the severity of antemortem cognitive decline. For example, Braak Stage IV pathology may combine with CVD to induce significant cognitive impairment, whereas Braak Stage V pathology may itself have more biologic impact, yet still be clinically silent in the absence of concomitant pathological abnormalities.

Concomitant Pathology: An Expanding List of Non-AD Brain Diseases in Aged Persons

Although not as prevalent as CVD, non-AD neurodegenerative diseases are important for consideration. Synucleinopathies are the second most prevalent neurodegenerative disease category after AD; these include dementia with Lewy bodies and Parkinson disease dementia. Depending on the study population, dementia with Lewy bodies and Parkinson disease dementia constitute approximately 5% to 15% of all neurodegenerative diseases (57, 88). Cortical Lewy body pathology occurs frequently in association with AD (88, 89). Whatever the reason for this clustering, the phenomenon is important from the perspective of CP correlation. When cortical Lewy bodies are present, the correlation between AD pathology and cognitive impairment becomes far less strong (63), probably because the Lewy body pathology is accounting for additional cognitive impairment. Accordingly, we and others have found that patients with AD and dementia with Lewy bodies decline at a faster rate than pure AD (90–92). The most reliable diagnostic tool for detection of neocortical Lewy bodies (i.e. immunohistochemistry for α-synuclein) was not developed until the late 1990s, so earlier studies had less ability to detect these structures (93, 94). There are now known non-AD dementias linked to mutations in valosin, progranulin, TDP-43, CHMP2B, and various triplet-repeat alleles (6, 95). These relatively newly discovered diseases present an important caveat to interpreting older studies that tackled the problem of CP correlation in dementia.

In addition to neurodegenerative diseases per se, some other common diseases such as hippocampal sclerosis, substance abuse, mood disorders, cancer, hematomas, and hydrocephalus are directly contributory to cognitive impairment in the elderly (96–98). Together, neurodegenerative diseases, CVD, and other aging-associated brain diseases weaken the apparent association between AD pathology and antemortem cognitive impairment. An analogy can be made to studying heart disease: it would be difficult to study the CP correlation between coronary atheromas and heart function if three quarters of cases had bacterial endocarditis and/or severe arrhythmias as well. That would be true even if clinical parameters could be correctly and confidently assessed in each case.

Challenges Pertaining to Cognitive Assessment

Difficulties in CP correlation may have as much to do with the clinical as to the pathological component. In AD-relevant CP studies, the clinical aspect is commonly the severity of cognitive impairment as quantified by test scores. First, it is problematic to align cognition and its many substituent functions on a linear scale. Second, there is added difficulty in correlating brain disease with test scores. In other words, even if cognition could be placed on a valid linear continuum (a test score), it would still be unsafe to assume that regression along that continuum will occur in a linear fashion that correlates with brain disease severity, whether a histopathologic substrate exists.

Although outside the scope of this review, specific cognitive assessments are used for differing applications. More challenging, memory-oriented, and sensitive tests such as the AD Assessment Scale, Cognitive component have been used frequently in clinical trials and clinical imaging studies when intraindividual assessments can be correlated as an outcome measure. By contrast, most CP studies have relied on less sensitive tests such as MMSE, the Blessed Test, or the even more semiquantitative Clinical Dementia Rating Scale for CP correlation (26, 36, 52, 61, 63, 90, 99). These latter tests have been more constantly used over time, assess multiple cognitive domains, have relatively well-established normative values, and could be more specific for interpersonal and cross-study CP correlations. It may be best to use a comprehensive battery of cognitive and functional tests such as those used currently in the Uniform Data Set by Alzheimer’s Disease Centers (100).

A further complication to CP studies is that many nonstructural factors alter cognitive assessment in the elderly. These include cardiovascular, metabolic, infectious, medication related, and mood disorders and debilities such as fatigue, arthritis, hearing loss, and visual impairments that can also change test-taking behavior.

Collectively, many clinical factors render inevitable some degree of variation in the results of cognitive assessments and hence in correlating test scores with any other parameter. There are thus many layers of complexity to CP studies of AD. Yet, by taking into account some of the above challenges, we can begin to address the existing CP literature and evaluate the hypothesis that amyloid plaques and NFTs contribute in a biologic sense to the clinical manifestations of AD.

Correlation of Amyloid Plaques and Cognitive Decline

More than 30 CP studies have assessed correlations between the density of amyloid plaques in human brains and the severity of antemortem cognitive decline (26, 34, 36, 56, 61, 63, 99, 101–106). These studies varied in research cohorts, anatomical areas examined, pathological methods, amyloid plaque subcategories, plaque-counting techniques, metrics for cognition and the range of cognitive decline tested, and the rigor with which concomitant pathologies were evaluated and/or factored into the study. Controversy exists about the best way to calculate the extent of amyloid plaque pathology. Neuritic plaques can be visualized via a variety of techniques, including silver stains, thioflavine S, and immunohistochemistry. Excellent studies have put forth differing views of whether one should use neuritic plaques, diffuse plaques, total plaque load, amyloid burden, or other pathology metrics (107–109). Studies also use differing techniques for counting and/or scoring their density (107).

Despite the wide variations in study designs, several points emerge consistently among CP studies about amyloid plaques. First, compared with NFTs, there is a weaker direct correlation between the density of amyloid plaques and the severity of cognitive decline. Second, the amyloid plaque subtype that seems to correlate best with the severity of cognitive decline is the neuritic plaque. Third, the patterns seen in aged persons’ brain seem to segregate into these 3 groups (note that many refers to counted lesions in neocortex): 1) Few plaques, few NFTs, and no cognitive impairment; 2) Many plaques, few NFTs, and no cognitive impairment; and 3) many plaques, many NFTs and cognitive impairment.

Because in part of the scarcity of neocortical NFTs in the absence of plaques, it has been suggested that the pathogenetic effect of amyloid plaques may be mediated through stimulating neurofibrillary pathology (63, 110). The relative importance of amyloid plaques in terms of correlating with cognitive decline may be greatest in the earliest stages of the disease, before there are widespread NFTs and neocortical neurodegeneration (99). By contrast, there is relatively little evidence of a direct effect of amyloid plaques in contributing to the later-stage cognitive decline in AD. Germane to this topic but often neglected is the question of lesion turnover. The apparent success of Aβ vaccine trials attests to the ability of the immune system to clear amyloid plaques (111), but it is not known whether a toxic plaque substance is left unaffected in these studies. Microglia appear to actively scavenge the fibrillar material in amyloid plaques (112). There is more direct evidence of amyloid plaque turnover in animal models (113). This phenomenon would tend to dampen the correlation between pathology and clinical features.

To a degree far greater than for NFTs, there is a strong association between amyloid plaques and genetic factors of risk for AD. As a rule, environmental and genetic factors that predict AD risk also appear to directly potentiate amyloid plaques. These risk factors include head trauma, the ApoE4 allele, Down syndrome, amyloid precursor protein mutations and duplications, SORL1 variants, and mutations in PSEN1 and PSEN2 genes (13). This strong genetic association is relevant to the question of the correlation with cognitive symptoms because all of these alleles are also strong risk factors for dementia.

The Neuritic Plaque: A Pathological Entity With Strong Mechanistic Implications

Neuritic plaques deserve special consideration. Neuritic plaques comprise roughly spherical extracellular amyloid deposits that are invested by degenerating or dying back nerve cell processes (Table 1; Fig. 1). These abnormal dendrites and axons contain aberrant tau fibrils identical to those seen in NFTs. In a given neuritic plaque, axons from a variety of different sources expressing distinct neurotransmitter signatures may be present (114, 115). This is therefore an important clue to AD pathogenesis because neuritic plaques represent unambiguously a nidus in which extracellular amyloid plaque pathology induces intracellular neurofibrillary pathology and apparent structural and functional disruption (116, 117). Although the particular toxic substance(s) within neuritic plaques are still not definitively known (i.e. they may represent specific conformation(s) of the Aβ peptide [118]), this apparent “smoking gun” is difficult to interpret in any other way.