A History of Senile Plaques: From Alzheimer to Amyloid Imaging

Abstract

Senile plaques have been studied in postmortem brains for more than 120 years and the resultant knowledge has not only helped us understand the etiology and pathogenesis of Alzheimer disease (AD), but has also pointed to possible modes of prevention and treatment. Within the last 15 years, it has become possible to image plaques in living subjects. This is arguably the single greatest advance in AD research since the identification of the Aβ peptide as the major plaque constituent. The limitations and potentialities of amyloid imaging are still not completely clear but are perhaps best glimpsed through the perspective gained from the accumulated postmortem histological studies. The basic morphological classification of plaques into neuritic, cored and diffuse has been supplemented by sophisticated immunohistochemical and biochemical analyses and increasingly detailed mapping of plaque brain distribution. Changes in plaque classification and staging have in turn contributed to changes in the definition and diagnostic criteria for AD. All of this information continues to be tested by clinicopathological correlations and it is through the insights thereby gained that we will best be able to employ the powerful tool of amyloid imaging.

INTRODUCTION

This review focuses on the history, histological classification, brain distribution, and clinical significance of senile plaques as well as how these have affected diagnostic criteria for Alzheimer disease (AD). The intention is to provide a concise summary of the accumulated knowledge that will help in the understanding and employment of both histological studies and amyloid imaging as they relate to AD research, diagnosis, and clinical trials. For discussions of more general issues in the history of AD, including the evolving understanding of the relationships of senile plaques with neurofibrillary tangles, or of the detailed genetic, molecular, biochemical, or etiological data related to senile plaques or AD, or of recent progress in amyloid imaging, other sources might be more suitable (1–21).

THE EARLY YEARS: RAPID DESCRIPTION OF MANY PLAQUE CHARACTERISTICS

As for many advances in science, the discovery of plaques and the subsequent early burst of characterization studies were facilitated by methodological innovations. In the 1890s, the silver stains of Golgi and Cajal illuminated the structure of the nervous system as never before. Blocq and Marinesco, working at the Salpêtrière in Paris and using hematoxylin and eosin as well as their own modifications of Cajal’s methods, are usually credited with the first description of senile plaques (12, 15). They described them as “round heaps” and suggested that they were glial scars but did not appear to have commented on their possible significance. In Vienna, Redlich, using the dye Carmine Red, described plaques in 2 more cases in 1898, calling them “miliare Sklerose” or, translated to English, “miliary sclerosis,” after their resemblance in size, it is presumed, to millet seeds scattered in the brain. He also referred to these as “plaques,” probably the first use of the term (15). In publications from 1902 and 1903 (15), Bielschowsky contributed the ammoniacal reduced silver method that now bears his name and used it to demonstrate threadlike structures in normal neurons that he called neurofibrils. Alzheimer used it 3 years later to stain not only plaques but to also give the first description of neurofibrillary tangles in his patient Auguste D (10). Others quickly followed and by the end of 1912 more than 45 articles had been published from studies of more than 500 brains, describing plaques and tangles in both demented and nondemented middle-aged and older persons, as well as a very few younger people. Thorough case reports and literature reviews indicate that these early studies quickly established an astonishingly comprehensive sketch of the great variations in plaque structure as well as their regional brain distributions and age and disease associations.

Writing in the period between 1910 and 1913, Solomon Carter Fuller (Fig. 1) at the Westborough State Hospital in Massachusetts (22–24), and Albert M. Barrett of the State Psychiatric Hospital at the University of Michigan (25, 26), paint a lively and accessible (especially for those limited to the English language) picture of the progress of work issuing forth from a score of laboratories around the world. Fuller (1872–1953), perhaps the single most remarkable investigator in the history of AD (27), was born in Liberia, the grandson to an American slave who had bought his freedom and returned to Africa. Fuller’s father had done well, as the owner of a coffee plantation and a government official. At the age of 17, Solomon retraced his grandfather’s steps to America, attending college first in Salisbury, North Carolina and then on Long Island. He graduated with an MD in 1897 from Boston University School of Medicine and in 1903 was 1 of 5 foreign students chosen by Alzheimer to do research in his Munich laboratory. He joined what was then a powerhouse of neuropathology that hosted such luminaries as Bonfiglio, Creutzfeldt, Jakob, Lewy, Merzbacher, Perusini, and Simchowicz (8). Upon returning to the United States, in his 1912 major review (22), Fuller quotes translations of earlier German, Italian, and French studies, beginning with Redlich’s description of plaques, “Miliary plaques consist of numerous fine glial fibres…generally circular in outline…”. He concludes that they have dead or dying ganglion cells at their center and are surrounded by “…focal compensatory glial proliferation…”. Alzheimer, in 1904, had described plaques as a “…feltwork of fine glial fibres surrounding corpora amylacea or other bodies the nature of which was not clear.” The term “amyloid plaque” was not yet in usage however, as the relationship of plaques with amyloid stains had not yet been discovered. Alzheimer and others often referred to plaques as “Drusen” or “drusige Nekrosen.” “Druse” is translated to “geode” in English (15), and presumably it was the dense plaque core with its radiating character that inspired this name. Koichi Miyake of Tokyo University, who had trained in Vienna (15), in 1906 had used the term “glia Rosette.” These earliest reports emphasized the glial component of plaques. After the Bielschowsky method, with its specificity for neurofibrils, came into widespread use, the neuronal component was more confidently asserted, for example, from Fuller’s translated quotes of Fischer (24), then working out of Arnold Pick’s psychiatry department at the German University of Prague (15). In 1907, Fischer had described, “…club-like proliferations of axis cylinders surrounding the plaques.” In a second, larger study of 275 brains in 1910, Fischer found plaques in 56 cases, all of whom were over 50 years of age (15). Outside of the cerebral cortex, Fischer found plaques in the thalamus, striatum, and cerebellum but not in the brainstem or spinal cord. Fuller, still preferring the term miliary plaques but also referring to “so-called senile plaques,” reported on his own study of 89 brains (24), using not only the Bielschowsky method but also Mann’s eosin-methylene blue, acid fuchsin light green, Mallory’s phosphotungstic acid-hematoxylin glial stain, Herxheimer’s scarlet stain for fat, and Van Gieson’s stain. He describes plaques as having a “dark, circular, homogeneous nuclear-like mass [that] is centrally disposed…In the outer portion… glia and nervous elements and other not definitely determined structure[s]—fibrils, granules and globules are found…”. “Gliogenous phagocytic cells in frozen sections treated with Herxheimer’s stain for lipid substances… are frequently met with…”. Fuller confirms Fischer’s earlier descriptions (15, 24) of what have come to be known as dystrophic neurites, within what we would now call a neuritic plaque (Fig. 2), “A somewhat distinctive type of plaque is best shown in silver impregnated sections. Here, it will be seen, relatively coarse, tortuous curled and club-shaped fibres dominate the picture.” He suggests that these fibers are neuronal rather than glial, “These fine fibrils and by far the great majority of the coarse fibrils were not stained by any of the methods which display glia fibres…”. “Methods which display axis cylinders reveal these fibrils, although it must be admitted in a way which suggests that the supposed axis cylinders are undergoing a degenerative process.” Fuller also indicates the presence of morphological subtypes, perhaps related to stages of plaque development, “Plaques without a homogeneous nuclear-like mass are not rare…” and indicates that “…certain reactions of the glia… serve to differentiate what appear to be younger forms from old ones.”

FIGURE 1.

Solomon Fuller with Alois Alzheimer and colleagues, 1904, at the Neuropathology Institute, University of Munich. Fuller is seated in the center next to Alzheimer. Reprinted from Ref. (27), with permission from the Journal of History and Neuroscience, Taylor & Francis Group.

FIGURE 2.

Photomicrographs from Solomon Fuller’s 1912 publication in the Journal of Nervous and Mental Disease. These are probably from sections stained with the Bielschowsky silver method, although not indicated in the original publication. On the left is a lower-magnification view of “Right prefrontal cortex showing a rich deposit of plaques.” On the right is a “Typical plaque. Alzheimer degeneration [neurofibrillary tangles] also shown.” Reprinted from Ref. (22), with permission from Lippincott Williams & Wilkins.

Surprisingly, another American “foreign student,” Albert Moore Barrett, also trained in Alzheimer’s laboratory, but a few years earlier (1901–1902), when Alzheimer was in Heidelberg with Franz Nissl. Barrett graduated as Doctor of Medicine from the State University of Iowa in 1893. He was a physician and pathologist at the Iowa State Hospital for the Insane from 1895 to 1901. After training with Alzheimer, he was employed at the Danvers Insane Hospital, Massachusetts, and from 1905 to 1906 he was assistant in Neuropathology at the Harvard Medical School. In 1906, he started with the University of Michigan as Associate Professor of Neural Pathology and Director of the Psychopathic Ward of the University Hospital. He was also a pathologist for the Michigan State Asylums for the Insane (University of Michigan, Faculty History Project, https://quod.lib.umich.edu/b/bhlead/umich-bhl-9766?view=text). Barrett’s publication in the American Journal of Insanity in 1911 (25) provides a similarly detailed literature review as provided by Fuller, as well as the results of his own investigations. His findings parallel those of Fuller, giving accounts of what we would now call plaque cores and haloes, as well as their microglial and astrocytic reactions. “A Bielschowsky preparation shows… a solid black mass lying in the center of the plaque…[with]…altered neurofibrils…[although] …it is not easy to distinguish which… are nervous elements…or which are glial fibres…there are undoubtedly pathologically altered neurofibrils in the plaques.” He feels that “…the plaque [is] treated as a foreign body….”, culminating in glial encapsulation. Old plaques were encapsulated with glial fibers while if there was no marked gliosis, these were considered younger stages. As with Fuller, Barrett finds that there are often “…amoeboid types of glial cells containing fat…” within or around the plaques. Most startlingly, Barrett appears to describe what we would now call diffuse plaques and infers that they predate the other types, “The younger stages of the plaque have more extended areas. …Not infrequently they occur in the first stratum [first cortical layer]… these early stages show as focal areas… in which is accumulated a peculiar, finely granular or homogeneous substance. In the beginning this seems rather loosely spread.” “The beginning of a plaque seems to be a deposit of a peculiar substance in the fine cortical meshwork…”. “We found, in a number of instances, substances in every way reacting like the plaque substances, spread out in broad and long patches in the cortex. In one case [this] was spread through the first [cortical] layer, without any trace of the circumscribed limitation occurring in the [conventional] plaque.”

Both Fuller and Barrett reviewed what had been described on the brain regional distribution of plaques and again this is surprisingly congruent with our current understanding. Fuller quotes Redlich as finding that plaques showed a predilection for cortical pyramidal cell layers 3 and 5 while he himself at first finds plaques not related to “cortical stratigraphy” (22) and Barrett agrees with Redlich that plaques are more numerous in the “…upper nerve cell strata” (25). Fuller found plaques more frequent in the gyral sulci than on the gyral crests, and more frequent in the frontal and temporal cortex and hippocampus while being least frequent in the occipital cortex. Barrett thinks plaques are most numerous in the parietal lobe angular and supramarginal gyri, as well as the adjacent occipital convolutions. Again predating much later investigations, Fuller describes plaques in the lenticular nucleus, thalamus and brainstem but “…it is not clear to our minds why the cerebellum and spinal cord should exhibit a comparative immunity.” Barrett finds plaques in the striatum and thalamus as well as in the cerebellum, “although few in numbers” in the lattermost.

It is clear that this early spate of investigations also quickly came to modern conclusions regarding the clinical significance of plaques, although not without some dissension. A key issue was whether plaques were specific to dementia or not. Fischer (15), in a 1912 paper, found plaques in only 2 of 35 nondemented normal older people, which he put down to presymptomatic disease, while Miyake and Oppenheim, as reported by Fuller (22), found plaques more commonly in such subjects. Simchowicz, attributed by Goedert (15) as being perhaps the first to use the full term “senile plaque,” and who may also have been the first to use quantitative methods for counting plaques, stated after a study of 108 brains, according to Fuller (24) “Although sparse, plaques are found with fair regularity from the 80th and 90th year onward…” but “only in senile dementia are they in great abundance.” Fuller himself found no plaques in the brains of 50 young people.

Bielschowsky, who serendipitously catalyzed this early flowering, also contributed to the descriptions, proposing that both tangles and plaque neurites were caused by deposition of the same unknown substance, and classified plaques, as did Fischer, into up to 8 different types (15).

THE WORLD WARS ERA: A SLOWING OF THE PACE

From the sudden decrease of publication output alone, it is clear that the First World War had a catastrophic effect on German science, and, as Germany in particular had been so preeminent, this left a great global vacuum that would not be filled to the same extent for many decades. Thus, as late as 1930, the Scottish psychiatrists D.K. Henderson and Sarah MacLachlan (28), stated, “While considerable attention has been paid to Alzheimer’s disease in German, Italian and French literature, British psychiatry has taken little notice of it”. W.H. English, of the Rochester State Hospital in New York, writing in 1940 (29), says, “… it is only recently that a description of the pathological changes [of Alzheimer’s disease] has appeared in English textbooks of neuropathology.” Although the destruction of the European economy by the war must have been a major factor, there was also a deliberate academic boycott against German-speaking scientists (30). Countries associated with the Allies excluded German scientists and the German language from international associations, congresses, and publications. The boycott was only lifted in 1926, after Germany became a member of the League of Nations. Unfortunately, the 1930s brought new impediments to German medical research as a direct result of policies of the National Socialist regime (31). Oskar Fischer endured academic discrimination, finally being arrested by the Gestapo in 1941 and dying a year later in a political prison in Terezin, Bohemia (15).

The interwar years, with the Great Depression, followed by the Second World War, were thus marked by a much slower pace of investigation, but despite this there were several notable contributions to the characterization of senile plaques. Teofil Simchowicz continued and extended his quantitative approach to plaques, calculating frontal/occipital plaque density ratios to define subtypes of disease (32). Congo Red, a currently popular clothing dye, was found to bind to amyloid in 1922 (33) and in 1927 the Belgian neuropathologist Paul Divry used it to stain senile plaques, resulting in the first definite linkage of “amyloid” with “plaque. . . Cajal’s silver methods, which stimulated Bielschowsky’s own investigations, were further developed by Cajal himself and then del Rio-Hortega and Penfield (34) to classify neuroglia into astrocytes, oligodendroglia, and microglia. In 1940, English (29) used Bielschowsky’s and [von] Braunmuhl’s silver stains as well as Cajal’s new methods, finding that “…many [plaques] appeared to be surrounded by astrocytes.” Cajal himself had earlier investigated senile plaques (35), considering them examples of neuronal regeneration. After World War II, George Jervis (36) used the same set of stains that English had used, as well as the Holzer method for glia, on 3 cases of Down syndrome, confirming the earlier report by Struwe (5) that plaques were numerous in the cerebral cortex of affected persons, adding the observation that, “Proliferated glia cells were often seen to surround senile plaques.” Nils Gellerstedt in 1938 reached an approximation of the modern and profound conclusion that 80% of all people over age 65 had at least some senile plaques and neurofibrillary tangles (11).

THE 1960S: ULTRASTRUCTURAL PRECISION AND THE ARRIVAL OF STATISTICS

The 1960s dawned with the cellular source of plaques still murky. William McMenemey, addressing a meeting in Stockholm in 1961 (37), noted that some favored degenerating neurons as the primary process while “…some support the early view, upheld amongst others by Alzheimer that they [plaques] have their origin in the glial reticulum.” McMenemey presciently (or knowingly) mentions the “…interesting electron microscopic findings…” of “…my colleague, Dr. Michael Kidd…” and opines that, “I believe it will only be a matter of time before the electron microscope will solve the question of the siting of the plaque.”

Indeed, this was true, as both Kidd and Robert Terry were at that moment on the verge of answering this very question, working with biopsied cortex from AD patients. The electron microscope (EM), invented in the 1930s but only brought into commercial production in the 1950s, offered a degree of precision and confidence in cellular identification that was previously impossible. In the nervous system, neurons and the subtypes of glia could now be positively identified by the appearance and size of their characteristic cytoskeletal elements and organelles. Both Kidd and Terry noted the preponderance of altered neural elements within plaques. According to Kidd’s description of neurofibrillary tangles (38), “Most of the neurons were normal…. A number, however, were very different…[containing] thick bundles of parallel filaments… sweeping round the nucleus… At high magnifications the filaments composing the bundles were seen to be double helices” (perhaps inspired by the elucidation of the structure of DNA just a few years before by Watson and Crick). Kidd termed these paired helical filaments (PHFs). As recounted by Kidd and Terry in their historical reviews (9, 39), Terry had actually published first but had perceived them as twisted microtubules. Curiously, this may not yet be resolved as Pollanen et al (40) later published that aggregating tau filaments might actually form a twisted ribbon rather than PHFs. Kidd extended his report to plaques in 1964 (41), describing the plaque’s amyloid core, “Some, but not all plaques, showed large central masses of extracellular filaments which extended towards the periphery in rays of extracellular space” as well as the surrounding altered neurites, “The peripheral processes could be divided into those containing [paired] helices, those with many mitochondria, those with small lamellated bodies, and those with large lamellated and loculated dense bodies.” Terry, Gonatas, and Weiss (42) also noted that plaques (Fig. 3) contained “many enlarged axons and dendrites containing abnormal numbers of neurofilaments…” that were “…essentially similar…” to the neurofilaments within neurofibrillary tangles.” By 1967, Suzuki and Terry (43) stated more directly, “The perikarya of many neurons near the plaques contain neurofibrillary tangles. It is possible that these are the neurons whose distended dendrites and axons help to form the plaques.” Thus, between them, Kidd and Terry not only clarified neurons as the predominant cellular elements of plaques but for the first time convincingly linked plaques to tangles through their common constituent, PHFs. Also, and perhaps even more importantly, the nature of the substance within the plaque core was finally unambiguously recognized by both investigators, as stated by Suzuki and Terry, “…plaques have as their fundamental substance a … sort of filamentous material which is structurally identical to amyloid,” using as their proof recent published EM studies of amyloid elsewhere in the body. An experimental model of splenic amyloidosis was particularly contributory, and the envelopment and possible extravasation of amyloid fiber bundles by splenic reticuloendothelial cells led Terry to hypothesize that microglial cells, as the reticuloendothelial cells of the brain, may be responsible for the production of plaque amyloid. Both Kidd and Terry identified the presence of astrocytic processes at the plaque periphery. Extending structure to function, Gonatas (44) emphasized the presence of degenerating presynaptic terminals within plaques and realized that these “…may be significant in view of [their role] in learning and memory.”

FIGURE 3.

Electron micrograph of a senile plaque from Terry et al 1964. AM, amyloid fibrils; NP, neuronal process; DB, dense body; L, lipid. Reprinted from Ref. (42), copyright Elsevier.

The 1960s also saw the first serious usage of statistics in plaque research, as in 1962 Corsellis published his analysis of more than 200 brains of asylum patients, using t-tests, proportional tests, and correlation analysis to show a convincing separation of “organic” (including senile dementia cases) from “functional” conditions (45). Stimulated by this, Martin Roth, Bernard Tomlinson, and Gary Blessed undertook for the first time a statistical clinicopathological study (46, 47) wherein they would apply “…quantitative measures both to the phenomena of mental deterioration and to the pathological changes…”. Patients admitted to hospital wards, both psychiatric and general, including “control” subjects who were free of psychiatric symptoms, were rated with a “dementia score” for functional performance in “personal, domestic, and social activities” as well as a separate “test score” for memory and concentration. Twelve samples of cortex from the frontal, occipital, and temporal lobes were sectioned at 25 µm and stained with the von Braunmuhl silver stain. Plaques were counted in each sample in 5 random fields, the mean number of plaques per field determined and an interobserver correlation was calculated (0.98). A total of 264 subjects received the Blessed scales, 92 came to autopsy and 76 subjects had plaque counts done while 16 were excluded because of obvious ischemic lesions. Analysis of all 60 included subjects showed a highly significant correlation of both the dementia score and the test score with plaque count (0.77 and −0.59, respectively, both with p < 0.001). Among those with clinically severe dementia, though, the correlations were weak and not significant (0.11 and 0.22, respectively, for dementia score and test score). This study is widely regarded as the first definitive evidence that plaques had a dose-response relationship with functional performance and became the template for many more such investigations, and the Blessed Dementia Scale was widely adopted. However, the relatively weak correlation for degree of impairment, within the group with severe dementia, was to become a continuing source of debate on the clinical significance of plaques.

The 1960s closed with intriguing mysteries remaining for McMenemey to ponder on (37), such as why plaques prefer cortical association areas, certain cortical laminae, “…convolutional valleys rather than hilltops…” and why they occasionally “… clustered around a vascular bed…”. But plaque disease associations had been more fully ascertained, with foreshadowing of future related research fields, with further musings by McMenemey. “Argyrophilic plaques and neurofibrillary degeneration have been found together in post-traumatic dementia [and] in dementia pugilistica….” and, “Trauma certainly seems to be one such [predisposing] factor.” After summarizing previously reported cases of plaques, tangles, and dementia occurring years after both single and repetitive incidences of head trauma, he saw the future holy grail of AD prevention, “it must be our aim to recognize Alzheimer’s disease in the stage before the plaque….”.

THE 1970S: NEURITIC PLAQUES AND THE CORTICAL CHOLINERGIC DEFICIT

The initial electron microscopic characterization of plaques allowed Henryk Wisniewski and Robert Terry the perspective needed for a morphological classification, first published in 1973 (48), that heavily influenced researchers over the coming decades. “Many varieties of plaques have been described but in general the following types are recognized: typical or classical senile plaques are those with a central core of amyloid surrounded by granular and filamentous rods and granules and reactive cells; primitive or atypical plaques without the central core; compact or burned-out plaques are lesions made up of the central core alone.” “Since the degenerating or dystrophic neurites [axons and dendrites] are the major constituents of the plaque, and since we believe they start [the] formation of that complex structure, we propose to change its name from senile plaque to neuritic plaque.” Furthermore, they proposed that the 3 subtypes of plaque also represented their temporal evolution and that the degenerating neurites were the primary process, rather than amyloid formation. “Discarding amyloid as the initiator of the plaque…aggregates of degenerating neurites induce the local deposition of amyloid, and therefore neurites are the nidus of plaque formation.” “Microglia presumably respond to the degenerating neuronal processes, phagocytosing their debris and simultaneously depositing amyloid.” The burned-out or compact plaque was the end result, after the degenerating neurites had all been consumed, although they kept their options open, “… or may come from the primary deposition of amyloid.” Neuritic plaques could be further subdivided into “classical plaques,” with both a compact amyloid core and a halo of dystrophic neurites, “primitive plaques,” with only dystrophic neurites and no core, and “very primitive plaques,” with very poorly formed, small neuritic aggregates.

Admitting a heavy influence from Divry (49) and Schwartz, Terry (9) recognized that plaques were made up, at least partially, of amyloid, and so started using thioflavin S to stain plaques. “Silver stains were slow, expensive and of variable quality, so I turned to the thioflavin S following the work of Schwartz who had used thioflavin T to demonstrate senile amyloid in aged organs other than brain.” Terry was pleased as, “…the images were brilliant and the technique was sensitive, simple, quick and inexpensive.” A puzzling finding was that tangles also were intensely stained, despite their differing ultrastructural qualities.

Meanwhile, Taihei Miyakawa’s EM studies had suggested that amyloid originated from capillary basement membranes (50) and, in later publications suggested that virtually all plaques were in contact with, and hence probably produced by the vascular basement membrane (51, 52). The finding that some plaques appeared to grow out of the side of blood vessels had been described much earlier by Morel (53), and had been termed “dyshoric amyloid angiopathy,” but Miyakawa’s suggestion that all plaques were connected to blood vessels was a bold statement. In opposition to this, Friede had earlier reported contrary findings, with 92% of senile plaques unrelated to capillaries (54). Miyakawa’s use of serial sections may have given him a methodological edge, enabling him to see plaque-vessel contacts that were only in the section plane of a minority of single sections. In the following decades, several statistically based studies presented evidence that was mostly against the tendency for plaques to be always associated with blood vessels, but there is still no clear consensus (55–61).

In 1976 and 1977, 3 labs (62–64), headed by Peter Davies, Elaine and Robert Perry, and David Bowen, reported nearly simultaneously the presence of a depletion of cortical cholinergic markers in AD. This was tremendously exciting as it immediately raised the possibility that AD, like Parkinson disease, was functionally due to the relative lack of a neurotransmitter which, when replaced, would significantly alleviate disease symptoms. The severity of the cholinergic deficit correlated not only with decreases in a “mental test score” but also with increases in plaque counts (65). Some of the presynaptic boutons in the plaque contained cholinergic enzymes (66, 67). Further supporting a neuronal origin of plaques, Probst and coworkers confirmed, using Golgi preparations, Terry’s earlier ultrastructurally-based speculations that the dendrites of nearby neurons contributed to the dystrophic neurites within plaques (68).

THE 1980S: AMINO ACID SEQUENCING, IMMUNOSTAINING, AND DIFFUSE PLAQUES

The greatest single advance in the understanding of AD was undoubtedly the isolation and amino acid sequencing of plaque amyloid, as it enabled the identification of the nucleotide sequence of its precursor protein (β-amyloid precursor protein or β-APP) and chromosomal location. Glenner and Wong (69) are usually credited with being the first to publish an amino acid sequence, using meningeal vascular amyloid from AD and Down syndrome subjects, but David Allsop, working with Kidd and Landon, had in fact put out an 18 amino acid sequence a year earlier working with AD tissue (70). The interesting background to this, and the rivalries, are discussed in the recollections of Colin Masters and Konrad Beyreuther (6). Allsop’s group had deduced the partial sequence but not the N-terminal portion while Glenner and Wong identified 24 amino acids beginning at the peptide’s N-terminus. Masters and Beyreuther used plaque cores from AD patients and extended the identified N-terminal sequence to 28 amino acids (71) and predicted a total length of 40, while Alex Roher and colleagues identified a sequence similar to that presented by Allsop, Landon, and Kidd and deduced a theoretical total peptide length of between 36 and 40 amino acids (72). Glenner named it “β-amyloid protein” after the presumed β-pleated sheet conformation common to all amyloid, while Masters and Beyreuther (6) called it “A4” (the peptide’s name was later contracted by most to Aβ, which will be used for simplicity from here on). Discovery of the DNA coding sequence for β-APP and its localization to chromosome 21 by Goldgaber and collaborators (20) suddenly explained the plaques found in Down syndrome and at the same time made Down syndrome an even more insightful model for AD. Autopsy studies across the Down patients’ lifespans had indicated that plaques first appeared by age 20 or even earlier while cognitive deterioration was not evident until about age 45, suggesting a 20 or even 30 year preclinical interval for plaque occurrence and spread (5, 73). A long preclinical period was also suggested by the report that amino acids in the Aβ extracted from plaque cores were highly racemized (74). John Hardy’s group crowned this period with their finding of a missense mutation in the β-APP gene in an autosomal dominant early onset AD family and the articulation of the “amyloid cascade hypothesis” (13).

Knowledge of the amino acid sequence also made possible the generation of Aβ-specific antibodies (75) and their immunohistochemical (IHC) application sharply increased both the sensitivity and specificity of plaque detection, compared to the standard silver or Congo Red methods. Rozemuller and colleagues (76) summarized the aggregate findings of multiple groups of “a new type of plaque…the amorphous plaque” which were “…non-congophilic plaques that could not be visualized by the silver techniques of Bodian or Bielschowsky…”. Henryk Wisniewski (77) extolled the “…large areas of diffuse infiltration of the neuropil,” including the ribbon-like infiltration of the subpial layer of the cerebral cortex, and of, “…diffuse deposits in the molecular layer of the cerebellum and the basal ganglia…”. Tagliavini, Frangione, Verga, and coworkers (78, 79) described similar Congo Red and thioflavin-negative deposits which they called “preamyloid,” marked ultrastructurally only by electron-dense “flaky” deposits, while Yamaguchi and colleagues (80–85) called them “diffuse plaques” (which eventually became the most-used term) and found that they also contained “…some scattered bundles of amyloid fibrils…”. It is possible that the “amorphous plaques” described in 1983 by Gibson (86) with the von Braunmuhl silver stain and in the 1986 EM report of Allsop, Kidd et al (87) predated the IHC studies of diffuse plaques, and it even seems likely that Barrett, more than 70 years earlier, had clearly described them, especially the extensive subpial molecular layer deposits (26). Later work by others showed that although they were Congo Red-negative, at least a fraction of diffuse plaques could be visualized by Bielschowsky and thioflavin S, depending on the particular modification used. New, “enhanced” silver stains were also being developed to stain the new plaque type, including methenamine silver (80, 88, 89), Campbell-Switzer (90), and Reusche (91) methods. Braak and coworkers, using Campbell-Switzer, found large “clouds” or “lakes” of diffuse amyloid in the presubicular and parasubicular parvopyramidal cell islands (92) that were simultaneously described with both an enhanced Bielschowsky method as well as Aβ IHC by Akiyama et al (93). Braak, Braak, and Kalus showed that the Campbell-Switzer technique was as sensitive as Aβ immunostaining for diffuse plaques (94–97) while others compared Campbell-Switzer, Reusche, Garvey, modifications of the Bielschowsky stain and thioflavin S methods, finding them equal or even superior to Aβ IHC (93, 98–100). However, Aβ IHC, in most laboratories, offered superior reliability and sensitivity for plaques of all types (101) as compared to the classical stains, most likely because IHC is generally easier to perform than silver stains, and thus there was a resultant explosion of studies that showed that plaques were much more common and much more widely distributed throughout the brain than was ever previously realized (Fig. 4), as stated by Wisniewski et al (77), “…the extent of brain amyloidosis in Alzheimer’s disease is much wider than so far appreciated.” Foreshadowing the later studies of Braak and Thal (102), Ogomori et al (103) found the brain-wide distribution of Aβ deposits could be used to differentiate nondemented elderly subjects from those with AD and dementia, with deposits restricted mostly to the cerebral cortex in those without dementia while in AD dementia cases plaque presence extended to the hippocampus, amygdala, thalamus, striatum, claustrum, hypothalamus, midbrain, pons, and cerebellar cortex. See Figure 5 for examples of plaque staining by the author’s laboratory, using modifications of the Gallyas (104), Campbell-Switzer (90), thioflavin S (100), Bielschowsky (105), and IHC Aβ (106) stains.

FIGURE 4.

Reproduced photomicrographs from Wisniewski et al (1989), showing various patterns of immunohistochemical staining for Aβ in AD subjects. (A) Ribbon-like subpial diffuse Aβ deposits in the temporal cortex. (B) Higher magnification of temporal cortex diffuse deposits. (C) Diffuse Aβ deposits filling a presubicular parvopyramidal cell island. (D) Primitive plaque in subiculum, “trapping” a normal neuron. (E) Diffuse perivascular Aβ deposits, temporal cortex. (F) Small, compact Aβ deposits in the temporal cortex molecular layer. Reprinted from Ref. (77), with permission from Acta Neuropathologica, copyright Elsevier.

FIGURE 5.

Photomicrographs of senile plaques from different brain regions, stained with a variety of methods from the author’s laboratory, including the Civin Laboratory modification of the Bielschowsky silver stain (A, B), thioflavin S (C), immunoperoxidase for Aβ with the 6E10 monoclonal antibody (D), Gallyas silver stain (E), and the Campbell-Switzer silver stain (F–L). Neuritic plaques are seen in (A) and (E), diffuse plaques in (B), and both types in (C) and (D). The Campbell-Switzer stain also shows plaques in the frontal cortex (F), putamen (G), cerebellar cortex (H), hippocampus (I), presubiculum and parasubiculum (J), entorhinal area (K) and calcarine cortex (L).

Diffuse plaques did not have enlarged neuritic processes within them and therefore could not be called “neuritic.” Plaques could now be potentially classified as diffuse, very diffuse, neuritic, and compact (amyloid core only), and neuritic plaques could be subclassified as very primitive, primitive, and classical (core and halo). In fact, the widespread use of IHC was now allowing even further dissection of the neuritic subdivision. Identification of tau as the primary constituent of tangles had been accomplished almost contemporaneously with that of Aβ and amyloid, as described in the historical review by Jean-Pierre Brion (19), and soon afterwards tau, and especially abnormally phosphorylated tau (p-tau) was also found to be present within dystrophic neurites of neuritic plaques as well as densely filling the neuropil (in advanced AD cases) as “neuropil threads” (107). Critically, however, not all neuritic plaques had neurites identifiable with anti-PHF or anti-tau antibodies. This lack was particularly true for nondemented subjects (108–113), suggesting that PHFs and phosphorylated tau were later changes. Rather than PHFs or tau, some neuritic plaques had neurites that contained ubiquitin (114), cholinergic markers (67, 111, 115), chromogranin A (116), or other neurotransmitters or markers (117, 118), as reviewed by Dickson (21).

Another differentiating feature of diffuse plaques is that they did not appear to stimulate the microglial and astrocytic reactions characteristic of neuritic and cored plaques (119). Along with a lack of a localized synaptic loss (120), and their wide cortical distribution and density in nonimpaired older people, this suggested that they were relatively “benign” compared to neuritic plaques and this was generally borne out in clinicopathological correlation studies. The increased usage of IHC to study glial reactions, however, showed that both the astrocytic and microglial reaction to neuritic and cored plaques was much more vigorous than ever appreciated with classical staining methods. For microglia, the pioneering works of Rozemuller, Eikelenbloom, and collaborators (121, 122), the McGeer lab (123), the Rogers lab (124, 125), and Haga, Akai and Ishii (126) showed that “activated” microglia were consistently present within neuritic plaques. The continuing studies of the McGeers, in particular, coalesced into the “inflammatory hypothesis” of AD, wherein a “vicious circle” of positive feedback would start with the microglial reaction to amyloid, resulting in “bystander” damage to nearby healthy neurons, leading to their demise with resultant increased amyloid deposition and/or microglial activation, as articulated by Akiyama et al (127). The astrocytic reaction, as suggested by earlier studies, was more peripheral in the plaque (128). Wisniewski, Wegiel, and colleagues suggested, that as microglia surrounding plaque cores often contained ultrastructurally identifiable amyloid fibers, they were most likely secreting it (129, 130). A decade later, however, Akiyama and others favored glial phagocytosis and destruction of amyloid (131, 132) and thus the numbers of plaques might not be stable but would tend to “disappear” over time (133), as suggested earlier by Brun and Englund (134), culminating in an end-stage of AD with fewer, rather than more plaques.

The 1980s also saw the initial elaboration of the “neuroanatomical spread” hypothesis, based on the specific brain regional, cortical cytoarchitectonic, and laminar distributions of both plaques and tangles. Perhaps first clearly enunciated by Pearson et al (135), the preferential distribution of plaques to cortical association areas, and to cortical layers II and III, suggested that they could be spreading via cortico-cortical projections of the supragranular pyramidal neurons. Duyckaerts, Delaere, and colleagues (136, 137) confirmed this plaque distribution, finding that the distribution of diffuse plaques, however, was different, and that only neuritic and cored plaques correlated with intellectual status while diffuse plaques did not (109). The astrocytic reaction to neuritic plaques, as sensitively outlined by IHC for glial fibrillary acidic protein, amplified the laminar pattern of the plaques (128), showing that in nondemented subjects, neuritic and cored plaques tended to be distributed to supragranular layers II and III while in demented subjects, a bilaminar pattern was present, with frequent plaques also present in the infragranular pyramidal cell layer V. This suggests an initial spread, beginning in nondemented subjects, through the layer III corticocortical axonal projections, and a later spread, in demented AD subjects, along subcortically projecting layer V pyramidal cell axons. Rudelli, Ambler, and Wisniewski studied diencephalic plaque distribution and concluded that strong cortical connections were a characteristic of the affected regions, including claustrum, striatum, and certain thalamic nuclei (138). Struble (67), Arendt (139), and others (140) were amongst several to suggest that plaques were formed at degenerating cholinergic axonal terminals originating in the nucleus basalis of Meynert, thus linking the neuroanatomical spread hypothesis with the cholinergic hypothesis. Saper and colleagues (141) also articulated the neuroanatomical spread hypothesis and presaged the later revival of the prion hypothesis as applied to AD, more exactly, “The apparent spread of a degenerative process from neuron to neuron across normal synaptic connections suggests that the disease may somehow involve the transfer of a substance between neurons.” “…neuritic plaques are seen in such areas as the striatum and thalamus that receive projections from the affected cortical neurons.” Armstrong and coworkers later suggested that plaques were distributed in clusters suggestive of the modular columns of the cerebral cortex (142). Columnar deposits were also reported by Akiyama et al (143) while others found differential plaque distributions to cortical sulci and gyri (144). Ultimately, the neuroanatomical spread hypothesis received more recent and spectacular support, at least in experimental animal models, by the reports of several groups of apparent transsynaptic spread of inoculated tau and synuclein aggregates (145–148). Inoculation of Aβ has caused plaque-like structures to form (149–151) but there is as yet no evidence of transsynaptic spread and there is as yet no direct confirmation of this process in humans (152), although both phosphorylated tau and α-synuclein may have been transferred from host to graft in some patients with Parkinson and Huntington diseases (153, 154).

THE 1990S: STAGING, GRADING, CLASSIFICATION, AND DIAGNOSIS

Staging of disease progression and grading of its severity are concepts deeply embedded in oncology. No patient with cancer would be treated without first determining the stage, by anatomical spread of disease within and then outside the primary site, and the grade, by the histological appearance of the neoplastic cells. And yet by 1990, no similar staging or grading systems had yet been widely accepted for AD.

Into this gap stepped Eva and Heiko Braak. With their background in classical cytoarchitectonics (155, 156), they had the methods and neuroanatomical knowledge to see in detail what others had glimpsed but not completely comprehended or pursued. While the distribution of plaques and tangles had been roughly sketched out over the previous decades, without an appreciation of the finer details of mesial temporal lobe anatomy, previous investigators had completely missed the invariable and thus probably earliest telencephalic brain region affected by tangles in AD, the transentorhinal area, interposed between entorhinal allocortex and temporal isocortex (157). Standard sections of the hippocampus had previously been taken at the coronal level of the lateral geniculate body, too far caudal to include the transentorhinal area, which is optimally seen at the level of the amygdala and hippocampal head. The Braaks’ continuing usage, in the cytoarchitectonic tradition, of extremely thick sections (up to 1000 microns), brought out laminar and regional patterns that were difficult or impossible to appreciate in the much thinner paraffin sections that had become the standard in medical laboratories. Baillarger, as early as 1840, had in fact used hand-cut fresh slices of cortex held up to the light, discerning the now-familiar 6 isocortical (or neocortical) layers (158). An additional advantage exploited by the Braaks was their application of new, “enhanced” silver stains. These were the Gallyas method (159), selective for neurofibrillary changes including tangles as well as plaque neurites, and its complement, the Campbell-Switzer technique (90), which is relatively selective for plaques, including diffuse plaques. These silver stains penetrate and label their targets through the entire thickness of even the very thick sections used by the Braaks, while the poor penetration of immunostains (to about 10–15 microns below the section surface) limits the usable section thickness.

In 1991, the Braaks published the first comprehensive histological staging system for AD (160), with separate schemata for plaques and tangles. As this review is focused on senile plaques, the reader is referred to other sources for a full description of the staging as it pertains to neurofibrillary changes. What was immediately evident was the opposing patterns. While tangles appeared first in the transentorhinal and entorhinal areas, followed by the hippocampus, and amygdala, plaques appeared first in the neocortex, particularly the basal parts of the frontal, temporal and occipital lobes (A stage), only later extending to the remainder of the neocortex and hippocampus (B and C stages). Extension to subcortical regions including the striatum, thalamus, subthalamic nucleus, red nucleus and cerebellum was mentioned, but not brought into the staging system at that time. Diffuse and neuritic plaques were recognized but were generally not separately considered.

In 1997, the Braaks’ staging system became the focus of an entire issue of the journal Neurobiology of Aging, when their gargantuan report on staging of 2661 brains across the entire adult lifespan (161) was commented on by multiple leading investigators. The most controversial findings were that, first, contrary to the prevailing opinion, and going against the now-dominant amyloid cascade hypothesis, tangles appeared to precede plaques in that substantial numbers of subjects had tangles but no plaques, “In general, initial neurofibrillary changes (stages I–II) develop in the absence of amyloid deposits…” and second, both lesions were extremely common if not universal in extreme old age. Over age 80, only 1% of subjects were free of tangles while over age 90, only 21% were free of plaques. Duyckaerts and Hauw (162) reanalyzed the data, with several assumptions, and pointed out that the “mid-prevalence” (when half the population was affected) of neurofibrillary change occurred roughly 27 years prior to that for plaques (age 47 vs 74), while the mid-prevalence for dementia was about age 90, almost 40 years after tangles and more than 25 years after plaques. These were startling revelations but the sheer numbers of subjects in the study were enough to give these new paradigms a firm foothold.

There were criticisms, some vociferous, of the study. Silverman, Wisniewski et al (163) found it odd that the Braaks had completely ignored all previous plaque classifications and had simply lumped all plaques together as “amyloid deposits.” Cummings (164), was adamant that the enhanced silver stains used were inferior to IHC, attributing the tangles-first conclusion to inadequate sensitivity for plaques, “The most likely explanation for Braak and Braak’s position that neurofibrillary pathology generally appears before plaque formation is that there are well-established differences in the sensitivity of stains for plaques and tangles and the present study sees NFTs well although missing a subset of plaques,” and “Diffuse amyloid plaques are systematically underrepresented in any neuropathological progression or correlative study that does not use formic acid pretreatment and antibodies to β-amyloid.” The Braaks responded with equal vehemence, showing the close correspondence, in adjacent sections of the images produced with the Campbell-Switzer method or with formic acid-pretreated Aβ IHC (94–97). “Apparently, immunostaining and this particular kind of silver-staining are equally sensitive.” As mentioned earlier, previous and subsequent studies tended to back up the Braaks’ contention that IHC was not always or necessarily more sensitive than all other methods (93, 98–100). Vallet et al (100) had reported that a modified thioflavin S protocol was more sensitive than Aβ IHC for plaque detection and Dickson showed that thioflavin S is suitable for staining diffuse plaques (21), while Rosenwald et al (99) rated Campbell-Switzer as better than IHC, and Gallyas as virtually identical to IHC. Therefore, considering all reports, it seems likely that IHC is closely matched by Campbell-Switzer, Gallyas, and thioflavin S, when all methods are performed optimally. The equivalency of Campbell-Switzer and IHC methods for Aβ has received very recent endorsement in an imaging-to-autopsy study of an amyloid imaging agent, florbetapir, where cortical standardized uptake value ratio (SUVr) correlated better with Campbell-Switzer-stained plaque density (Spearman rho = 0.76; p < 0.0001) than with area occupied by Aβ-immunoreactivity (Spearman rho = 0.71; p < 0.001) (165).

The Braaks’ great works were incomplete in that they lacked adequate standardized clinical cognitive information on their subjects, but many past studies had clearly indicated that both plaques and tangles were present in normal aging but were more frequent in manifest AD with dementia (11, 15, 166–171), suggesting a threshold of histopathological lesion density beyond which dementia would ensue. The determination of this threshold was considered of the utmost importance in that it would establish a diagnostic, albeit autopsy-based, “gold standard” for the diagnosis of AD. The concept of a threshold emphasized the urgent need for a grading system. Those subjects with lesions at or above the threshold could be termed AD, or at least their dementia could be attributed to AD while those below the threshold, if demented, must have had another cause. Braak’s staging system, while a great advance, had neglected grading in that there could be a wide range of neurofibrillary densities within any one stage. One difficulty, however, with establishing what exactly the threshold might be was the lack of standardized methods for determining plaque and tangle densities. An initial organized attempt had been made under the auspices of the National Institute on Aging (172) in conjunction with their 1984 establishment of a network of Alzheimer’s Disease Centers (ADCs). These criteria depended predominantly on neocortical plaque counts at or exceeding certain densities, as the presence of neocortical tangles, while supportive, was not a requirement. This stance appeared to be justified by the earlier work of Terry and Katzman on “plaque-only AD” (4, 173), with data showing that 30%–50% of demented AD cases over age 74 had no neocortical tangles. The establishment of AD diagnostic criteria that neglected tangles was objected to by some respected neuropathologists. Tomlinson wrote (174), “Nowhere in the criteria is there recognition of the certainty which attaches to the diagnosis when many neurofibrillary tangles accompany plaques, as they do in the majority of cases at any age,” and, “If it is a fact that 30% of older American cases of Alzheimer’s disease are devoid of neocortical tangles, and the incidence of such cases is similar in Europe, then European investigators are failing to report and investigate them…”. A practical difficulty with the Khachaturian criteria was that it was time-consuming to count plaques. Henryk Wisniewski and colleagues (101) received 104 replies from neuropathologists in a survey that indicated that only 25% or less were counting plaques and tangles as part of their diagnostic workup.

In fact, the Khachaturian criteria, devised in 1985, quickly became obsolete with the discovery of diffuse plaques shortly thereafter. Counting all the plaques would include diffuse plaques, which, aside from often having no discrete shapes or sizes, had a poor correlation with cognitive impairment. In response to this as well as to Tomlinson’s concerns and Wisniewski’s survey results, a set of US researchers came together under the banner, “The Consortium to Establish a Registry for Alzheimer’s Disease” or “CERAD” and established new diagnostic criteria for AD, removing the need to count plaques, replacing this with diagrammatic templates allowing semiquantitative grading (175). Actual counting of plaques and tangles, although intuitively seeming to be more accurate, in practice had more variability than the CERAD semiquantitative templates (176–180). For CERAD, diffuse plaques could be recorded but were not given diagnostic significance, due to uncertainties about this, as “Some workers claim that diffuse plaques are more commonly encountered in the nondemented elderly… whereas others maintain that diffuse plaques are those most common in the cortex in AD” (175). The criteria for AD dementia set threshold densities only for neocortical neuritic plaques, and tangles were again considered unnecessary. The full protocol did, however, require the semiquantitative assessment of tangles, and these and other accumulated data “…may allow refinement of diagnostic criteria” (181). For the sake of reliability and standardization as well as sensitivity, the usage of Bielschowsky silver or thioflavin S stains was recommended, although not absolutely required. Another aspect of the CERAD criteria, as with the Khachaturian criteria, was that the threshold thought to correspond to causation of dementia was raised for successively older age groups.

The new criteria were reviewed by Markesbery (182), who had mild disagreement based on the complete reliance on plaques again. “This approach to the diagnosis does not use the neurofibrillary tangle, and some neuropathologists will not agree with this” and “… I believe that it is important to examine these structures [tangles] closely…” and “Recently, Arriagada and coworkers (183) found that the severity of dementia was better correlated with the density of neocortical neurofibrillary tangles in the neocortex than with senile plaque density…”. Braak, Braak, and Mandelkow (184) were more emphatic, writing “Neuritic plaques are… ill-suited for diagnostic purposes” and “Non-demented elderly individuals frequently exhibit huge numbers of cortical amyloid deposits,” and “…it is extremely important that the diagnostic criteria for the recognition of AD be reevaluated.” However, the Braaks did not distinguish between neuritic and diffuse plaques, calling them all “amyloid deposits” and their dismissal of the specific clinical relevance of neuritic plaques ignored an already weighty amount of evidence to the contrary (46, 47, 109–113, 136, 185–189).

Although some were concerned that not all cases with neuritic plaque densities meeting criteria were demented, or who clinically seemed to have AD but on autopsy did not have enough plaques, in a conference discussion published in 1997 (179), Suzanne Mirra articulated the emerging concept that perhaps exact criteria were not possible but rather should be considered probabilistic, “The CERAD pathological criteria are not perfect [but may be used to] classify a patient … as possible or probable.”

Restlessness remained however, and in 1997 the NIA, together with the Reagan Institute, again convened an expert panel to devise new criteria. The reasons given were summarized in an editorial by Bradley Hyman and John Trojanowski (190). “The rationale for updating the Khachaturian criteria is that subsequent studies suggested they fell short of the goal of distinguishing controls from cases of Alzheimer’s disease. For example, numerous studies demonstrated individuals who met Khachaturian criteria in terms of senile plaque number but had no history of clinical symptoms of dementia. Secondly, the idea of changing diagnostic criteria based on the patient’s age has not been supported. Thirdly, the absence of assessment of neurofibrillary tangles in the current criteria stands in contrast to numerous studies that show that tangle number and distribution more closely parallel dementia than plaque number or distribution. Finally, there was a move away from setting absolute numerical criteria….” The new “NIA-Reagan” criteria admitted that only a probabilistic rather than exact correlation with dementia could be achieved, and thus 3 levels were established, assigning to each demented subject a “high,” “intermediate,” and “low” likelihood that AD was the likely cause for dementia. The CERAD templates for neuritic plaque grading were retained and the Braak neurofibrillary staging system was incorporated. High likelihood was defined by CERAD “frequent” neuritic plaques and Braak stage V or VI. Intermediate likelihood was CERAD “moderate” neuritic plaque density and Braak stage III or IV while low likelihood was CERAD “sparse” and Braak stage I or II.

Reactions to the new criteria were mixed but without the level of rancor aroused by the Khachaturian or CERAD criteria (see Neurobiology of Aging, volume 18, supplement 4 for multiple invited critiques). Perhaps the most telling critique was that of Geddes (191), who noted that many subjects could not be unambiguously assigned to one of the 3 probability levels because the plaque criteria did not always change in lockstep with the tangle criteria, thus, how to classify someone with CERAD frequent neuritic plaques but only Braak stage III or IV, or even Braak stage I or II? This problem was later echoed by others (192, 193) and would eventually come to be viewed as a major flaw, but the NIA-Reagan criteria were generally viewed as a significant improvement, and, most importantly, together the CERAD and Braak systems had given AD rudimentary staging and grading.

The 1990s also saw the extensive exploration of different biochemically defined forms of Aβ and their relevance for plaque types and distributions. Iwatsubo and colleagues used end-specific antibodies for Aβ with 40 or 42 amino acids, finding that all plaques were positive for Aβ42 while only about one-third were positive for Aβ40 (194). Diffuse plaques appeared to be exclusively positive for Aβ42 while other plaque types appeared to contain both Aβ40 and Aβ42, suggesting that, if diffuse plaques were the first morphological plaque type, as evolving thought was now suggesting, then Aβ42 must be the initial species deposited. Nakabayashi et al reported that Aβ42 was the only species present in the putamen (195), which contained mostly diffuse plaques. Others, including Akiyama et al (196), found that Aβ40, while less common in diffuse plaques than Aβ42, was not completely excluded, and Aβ40 was actually the more constant form in amyloidotic blood vessels. Funato, Ihara, and colleagues (197) supported the primacy of Aβ42 in their biochemical survey of Aβ throughout middle to old age, with Aβ42 being the first to accumulate in cerebral cortex of normal subjects, starting from about age 50 onwards. Elevated levels could be detected in some cases even in the absence of plaques. This lent support to the concept of preamyloid as established by the EM findings of Bugiani, Tagliavini, and Giaccone (198), and the finding, by Alex Roher, Yu-Min Kuo, and group, of elevated “water-soluble” Aβ in AD cortex, consisting of monomers as well as dimers, trimers, and oligomers (199, 200). Roher later led an examination of AD cortex samples with a preponderance of diffuse plaques, isolating a 17–42 Aβ fragment that they concluded was the major component (201). A distinguished group headed by Thomas Wisniewski confirmed and clarified these observations, using as their starting material cerebellar cortex from Down syndrome subjects, with almost exclusively diffuse plaques present (202). Again the 17–42 peptide was identified and it was suggested to be formed through the combined action of α- and γ-secretase cleavage of βAPP. Akiyama and colleagues documented the existence of similar truncated peptides within macrophages and glial cells, suggesting a lysosomal breakdown of full-length Aβ (131).

More evidence seemed to show that cerebellar diffuse plaques were different in many ways from those in the cerebral cortex, perhaps constituting a “very diffuse” form. Two groups (79, 203) confirmed that amyloid fibrils were only very rarely seen in EM preparations of cerebellar diffuse plaques, the former group stating that “… the majority… of cerebellar diffuse plaques…is preamyloid.” Dickson et al (114) noted that thioflavin S, while staining diffuse plaques in the cerebral cortex well, stained cerebellar diffuse plaques poorly, and Larner (204) confirmed that they were negative for Congo Red.

Continuing characterization of the neurites within neuritic plaques occurred. Wang, Munoz, and coworkers found chromogranin A and β-APP immunoreactivity to mark virtually all dystrophic neurites (116, 205), making them a more universal marker of neuritic plaques than antibodies against tau or phosphorylated tau. Both Munoz’ group (205) and Suenaga, with Dickson and others (206) reported that not all striatal plaques were diffuse as some contained dystrophic neurites immunoreactive for chromogranin A, β-APP or ubiquitin while only a minority stained for PHFs or tau. The latter group also reported that neurites around cerebellar compact amyloid plaques did not immunostain for tau but did stain for ubiquitin (207). Collectively these new data indicated that plaques were much more complicated than had been appreciated only 20 years earlier. The plaque classification advanced by Wisniewski and Terry now included not only diffuse plaques but multiple types of neuritic plaques, based on the morphology of their neurites as well as a tentative chemical and temporal evolution. Within diffuse plaques, some (e.g. cerebellar molecular layer diffuse plaques) might be termed “very diffuse,” as they, unlike diffuse plaques elsewhere, were unstained or very poorly stained by thioflavin S. In fact, the complexity generated by the development of new stains and especially IHC, had resulted in almost unlimited combinations of plaque staining qualities, as discussed in Dickson’s 1997 review (21).

The puzzling difference between the brain regional and subregional distributions of diffuse and neuritic plaques elicited a hypothesis from Bugiani, Tagliavini, and Giaccone (198), “A study of the distribution of preamyloid and amyloid deposits of senile plaques showed that preamyloid… involve[s] all the gray structures, from cortex down to anterior horns whereas senile [neuritic] plaques are confined to [only] some of them…”. “These findings emphasize that [neuritic] plaque morphogenesis largely depends on neurites vulnerable to paired helical filament-related degeneration and suggest that in regions with no vulnerable neurites, preamyloid deposits…cannot evolve into senile plaques.” A similar hypothesis was advanced by another group (140) based on the intricate laminar disposition of diffuse and neuritic plaques in AD primary visual cortex.

Dietmar Thal took on the painstaking task of finding orderly patterns of whole-brain amyloid distribution, thereby elaborating and expanding on the Braaks’ initial cortical amyloid stages (160). First publishing on the mesial temporal lobe alone (208), he later expanded his studies to the entire brain, resulting in a whole-brain staging scheme in 2002 (102). Thal studied 51 brains with a known wide range of cortical amyloid deposits from none to many, as well as a range of mental capacity from normal to demented. He again used the Campbell-Switzer and Gallyas silver methods favored by Braak and again referred to evidence that these were equal to IHC for the detection of plaques and tangles. Based on the frequency of occurrence of amyloid deposits in a great number of brain regions, he constructed 5 stages, suggesting an expansion with time and disease severity from an initial neocortically restricted origin (Phase 1) to allocortex such as the entorhinal and hippocampal CA1 region (Phase 2) and then to progressively more subcortical locations, including striatum and thalamus (Phase 3), upper brainstem (Phase 4) and finally cerebellar cortex and lower brainstem (Phase 5). Although standardized cognitive information was available for only 32 cases, it appeared that Phases 1, 2, and 3 were mainly found in nondemented elderly subjects while Phases 4 and 5 were mainly restricted to those with AD dementia. Importantly, the amyloid phases were highly correlated with Braak neurofibrillary stage, allowing a rough prediction of neurofibrillary stage from the amyloid phase. A limitation of the Thal amyloid staging system, as with the Braak neurofibrillary staging system, is that it lacks both qualitative and quantitative grading. Stages were distinguished only by the presence or absence of amyloid deposits, with no attempt at plaque density quantification, and no subdivision of plaques into diffuse and neuritic. The very similar pattern of early-stage amyloid deposition, between nondemented and demented subjects, led Thal to hypothesize (as had Braak and many others before him), that the development of AD histopathology in normal older people truly represented the same disease process and that AD begins and extends through a prolonged preclinical period. “Because early stages of AD-related Aβ or neurofibrillary pathology… appear to be early steps in the process leading finally to the pathologic picture of AD… it is tempting to speculate that non-demented cases exhibiting early stages of AD-related pathology represent preclinical stages of AD.” He extrapolated the significance of the histopathological progression to the treatment of AD, “…one could hypothesize that treatment of AD [would be] more successful the earlier the expansion of Aβ and NFT pathology is stopped….”. He also realized, however, that his system still required statistical validation of its correlative value against clinical status, “Although demented individuals show higher Aβ phases than non-demented individuals, it is still questionable whether there is a correlation between an increasing degree of dementia and Aβ phases. In our sample only single cases exhibit CDR scores of 0.5, 1 and 2, so we cannot answer this question at the moment.”

THE NEW CENTURY: HOPE IN STATISTICS AND STANDARDIZATION

What was admirable about the systematization efforts of the 90s was the repeated, even dogged attempts to define and standardize staging systems and postmortem diagnostic criteria for AD. What was disconcerting was that these criteria were based on “expert opinion” and lacked comprehensive validation by standardized clinical correlation. While statistical analysis of clinicopathological relationships had become much more common, most of the studies lacked sufficient power and did not use multivariable methods to determine the strength and significance of independent factors such as age, cocontributory lesions from the same disease (e.g. plaque effect divorced from tangle effect) and other comorbid diseases or lesions (e.g. cerebrovascular lesions, Lewy body pathology, white matter rarefaction, argyrophilic grains, and others), with due consideration to statistical confounders such as multicollinearity. Thus, while the conventional wisdom had evolved to regarding the clinical significance of tangles as greater than neuritic plaques and that of neuritic plaques as greater than that of diffuse plaques (209), conflicting conclusions were frequent and no studies could be considered definitive. Yet the approach and commitment to standardization, exemplified but not exclusive to CERAD and NIA-Reagan, led to increasingly powerful longitudinal datasets at many centers, including the NIA ADC’s database housed at the National Alzheimer’s Coordinating Center (NACC) (210).

The comprehensive review of Nelson et al (3) is referred to for a summary of the impressive statistically backed clinicopathological correlation studies undertaken in the post-2000 era (as well as some preceding work). The review concurs with the major conclusions of the preceding decades, most particularly, that while neurofibrillary tangles are the dominating correlate of cognitive impairment, neuritic plaques (but generally not diffuse plaques) are also a separate and statistically significant factor. Although universally present even in nondemented elderly, tangles are restricted to subcortical locations in these subjects. When tangles are widespread in the cerebral neocortex, dementia is almost always present. Deposits of Aβ, on the other hand, are present in the majority of subjects of advanced age but are not universal, although some small studies have found universality in centenarians (211). Two important studies by Hyman and colleagues were published after the Nelson review, both examining NACC data. The first (212) found that for those with Braak stage V or VI and frequent CERAD neuritic plaque densities (“high” probability of dementia according to the NIA-Reagan criteria), 91% had moderate or severe dementia; of those with Braak stage III or IV and a moderate CERAD neuritic plaque density (“intermediate” probability of dementia), about 50% were rated as at least mild dementia, while for those with no or low levels of plaques and tangles (“low” probability of dementia), most were cognitively normal. The second (213) used multivariable methods, finding that both neuritic plaques and neurofibrillary tangles were independent and significant predictors of a functional dementia measure, while arteriolosclerosis, amyloid angiopathy, and hippocampal sclerosis were also independent associates.

The presence of frequent coexisting brain diseases or lesions was most frequently ignored or unappreciated in preceding decades but after 1990 there was an increasing focus on these, particularly coexistent brain infarcts (214–223 ) and Lewy body disease (224–230) but also hippocampal sclerosis (231–236), progressive supranuclear palsy (237–239), circle of Willis atherosclerotic stenosis (240–243), cerebral white matter rarefaction (244–249), argyrophilic grains (250–254), and TDP-43 pathology (255, 256). Unequal segregation of any of these conditions with AD, plaques or tangles could bias clinicopathological correlations toward or away from significance. To resolve the contributions of multiple comorbidities requires large subject numbers but several groups have provided valuable analyses taking multiple comorbid pathologies into consideration (213, 257–260). The study of Nelson and colleagues (261) is an example of how large numbers of subjects, and careful neuropathological assessment, could be used to begin to sort out the independent contributions of specific lesions by, at the same time, taking into account the contributions of all other measured lesions. Forty-three clinical and neuropathological variables were assessed in 334 autopsied subjects and using such multivariable statistical methods, concluded that, even taking into account all other variables, both tangles and neuritic plaques significantly contributed to cognitive impairment in both AD and nondemented persons. Abundant neocortical Lewy bodies and TDP-43 pathology were also found to be significant contributors while diffuse plaques and argyrophilic grains were not significantly associated with cognitive performance. The meaning of plaques present in older people without detectable cognitive impairment was still not clear, particularly whether these represented preclinical AD or just normal or “pathological” aging (262).

Meanwhile, clinicopathological studies of subjects with genetic factors affecting AD showed that subjects with autosomal dominant AD due to mutations in the βAPP or presenilin genes generally had earlier onset and more rapid progression of plaques as well as a faster spread to subcortical regions (263). Carriage of the apoE-E4 allele also was shown, in a large Finnish study (264) to result in an earlier appearance of plaques, beginning as young as age 30, and higher plaque densities at all ages. For example, in the 50–59 year age bracket, 41% of E4 carriers had neocortical plaques compared to only 8% of noncarriers. These findings were later replicated by others (265).

Plaque-directed therapies began to emerge and go to clinical trials, most notably those involving “immunotherapy,” with either “active” immunization with fragments of Aβ, or “passive” immunization with monoclonal antibodies (or nonspecific IgG) (266). Although showing slight improvement on some analyses, the active vaccines were judged to have too high a rate of encephalopathy, thought to be due to meningoencephalitis or vasculitis and edema. Subsequent trials with passive agents did not, until very recently (267, 268), show promise for effects on brain Aβ or cognition. Postmortem studies of the brains of immunized subjects still have uncertain conclusions due to low subject numbers and uncertain pretreatment status regarding the density and distribution of plaques, but it has been suggested that at least some types of immunotherapy may have been effective at removing plaques in a patchy, discontinuous fashion (269). Also raised, however, was the possibility that the “released” Aβ may linger in the brain in elevated soluble concentrations or migrate to nearby blood vessels and become attached to their surfaces (270).

Doubts about the reliability of staging and grading systems continued to fester, however, and these coalesced into a series of studies by BrainNet Europe, a consortium of neuropathology laboratories (271–273). All were dedicated to improving interrater agreement in the assessment of neurofibrillary change and amyloid plaques, and were also committed to the assumption that the molecular specificity of IHC would be superior to conventional stains. The initial study was published in 2006 (271) and compared both Aβ IHC and p-tau IHC to the Bielschowsky and Gallyas stains, using tissue microarrays. This had limited conclusions but found the “acceptable or good” percentage of stained slides was 53% for Bielschowsky, 58% for Gallyas, 71% for Aβ IHC and 94% for ptau IHC. It was felt that there was some variability due to differences in the antibodies used, postmortem intervals, fixatives, and duration of fixation. The initial study was followed up by 3 more studies, including one for p-tau (272, 274, 275), using full-sized paraffin tissue blocks from a reduced number of contributing laboratories, and specifying the usage of only one antibody each for Aβ and p-tau. The results of the blinded individual assessments of p-tau IHC by 25 investigators were not encouraging, as while there was 96% agreement for Braak stage VI, there was only 68% agreement for stage III/IV and 33% agreement for I–II. Subsequent joint assessment and training improved the agreement. It was concluded that poor familiarity with the detailed neuroanatomy required for Braak staging may have been a major shortcoming and that “Neuropathologists need detailed instructions…” with the staging rules. In contrast, the studies dealing with Aβ IHC had 81% interrater agreement on assigning Thal amyloid phases.

American researchers were also not completely happy with the NIA-Reagan criteria. Nelson et al (192) echoed the comments of his Kentucky colleague, Jim Geddes (191) 13 years earlier, that NIA-Reagan, by specifying plaque and tangle combinations for only 3 of 28 possible combinations of CERAD neuritic plaque densities and Braak neurofibrillary stages, left neuropathologists uncertain as to how to classify the “off-diagonal” cases. Others (276, 277) also found great variability in how individual NIA-ADCs assigned NIA-Reagan stage even when they had exactly the same CERAD and Braak stages. But the main stimulus behind the revision of the NIA AD neuropathological criteria came from the long-overdue admission from clinical AD investigators that AD had a long preclinical stage and that treatment was best directed at that stage.

In 2011, 3 consensus panels, jointly sponsored by the NIA and the Alzheimer’s Association, had recommended that AD be defined and studied in terms of 3 clinically defined stages, preclinical, mild cognitive impairment, and dementia (278–280). This was a paradigm shift in that previous expert clinician panels had always considered the diagnosis of AD to require dementia while leaving the definition and even existence of a preclinical stage in limbo. The panel that most directly articulated this paradigm shift was that reported on by Sperling et al (280) and this is extensively quoted here to give the reader a direct window into their thought processes. “Although the field is already moving to earlier clinical trials at the stage of MCI, it is possible that similar to cardiac disease and cancer treatment, AD would be optimally treated before significant cognitive impairment, in the presymptomatic or preclinical stages of AD.” “To facilitate the possibility of future presymptomatic/preclinical treatment of AD, our working group, as well as the other 2 groups, felt it was important to define AD as encompassing the underlying pathophysiological disease process, as opposed to having AD connote only the clinical stages of the disease.” “We postulate that AD begins with a long asymptomatic period during which the pathophysiological process is progressing, and that individuals with biomarker evidence of [AD] are at increased risk for developing cognitive and behavioral impairment and progression to AD dementia…”. The extent to which biomarkers of [AD] predict a cognitively normal individual’s subsequent clinical course remains to be clarified, and we acknowledge that some of these individuals will never manifest clinical symptoms in their lifetime. Thus, it is critical to better define the preclinical stage of AD, to determine the factors that best predict the emergence of clinical impairment and progression to eventual AD dementia, and to reveal the biomarker profile that will identify individuals most likely to benefit from early intervention.”

The 2010 model of Jack et al (281), with parallel sigmoid curves representing the age-related appearance and rise of several biomarkers including amyloid (as detected by CSF and PET imaging) and tau or p-tau (detected in CSF) with respect to cognitive decline represented the current “best guess” as to how these rise over time and eventually reach a threshold that results in dementia. Jack et al cited and was obviously inspired by the very similar sets of curves drawn by Duyckaerts and Hauw in 1997 (162), using Braak’s data on plaque and tangle prevalence in 2661 human brains (161), and in 2004 by Ingelsson, Newell, Frosch, Hyman, and others (282), from their own postmortem studies of 116 AD and control subjects. The latter study, added, in addition to histological data for plaques and tangles, biochemical measures of Aβ, synaptic loss, and gliosis. Updated models from Jack et al (283) are more complex but have Aβ accumulation preceding tau/p-tau and the other markers, and preceding cognitive impairment by 10 years or more. The NIA panel, however, left its final recommendations as explicitly experimental, “These recommendations are solely intended for research purposes and do not have any clinical implications at this time. It is hoped that these recommendations will provide a common rubric to advance the study of preclinical AD, and ultimately, aid the field in moving toward earlier intervention at a stage of AD when some disease-modifying therapies may be most efficacious.”

All this, however, was in direct conflict with the NIA-Reagan definition of AD, which stated that dementia must be present to qualify for the diagnosis. As a result, the NIA convened another panel to reformulate the neuropathological definition of AD, with recommendations published in 2012 (212, 284). The reasons for the reformulation were stated as, “From the clinical perspective, the concept of AD has evidence to include patients with milder symptoms, including the proposition that there is a preclinical phase of the illness… Moreover, data have accumulated demonstrating that some older individuals who were cognitively intact proximal to death had significant AD neuropathologic change…There is consensus to disentangle the clinicopathologic term Alzheimer’s disease from AD neuropathologic change…recent NIA-AA sponsored consensus reports on three defined stages in a clinical continuum that includes preclinical, mild cognitive impairment, and dementia.” The neuropathologists’ panel recommended the combined usage of Braak neurofibrillary staging, Thal amyloid phasing and CERAD cortical neuritic plaque density assessments, which considered together place each autopsied individual into one of 18 boxes in a table, similarly to NIA-Reagan, except this time, each box signified, rather than a low, intermediate or high probability that dementia was due to AD, a low, intermediate or high level of AD neuropathologic change (or “not” for no change). Also, to avoid the ambiguity of NIA-Reagan assignments, all “off-diagonal” boxes were explicitly assigned to one of the final categories. Preferred histological methods were IHC for Aβ and tau or p-tau but other “acceptable” methods included thioflavin S or “sensitive silver stains” (citing the work of Braak). It was noted, in fact, that IHC methods were not adequate for identifying neuritic plaques, “Preferred methods for detection of neuritic processes in senile plaques are thioflavin S and Bielschowsky, immunohistochemical stains will identify specific and partially overlapping subtypes… of dystrophic neurites that may differ in disease relevance.” Instructions were given on the clinical significance of the “neuropathologic change” levels, with “Intermediate” or “High” levels considered “…adequate explanation…” for the presence of cognitive impairment or dementia. Differently from NIA-Reagan, where nondemented subjects were rated simply as “Criteria Not Met” and were not assigned a level, “We recommend that neuropathologists adopt the term AD neuropathologic change and report on its presence and extent regardless of an individual’s cognitive state.” As with NIA-Reagan, other possible causes of cognitive impairment were to be sought out, and the reports gave an outline on how to do this for Lewy body disease, “vascular brain injury” and hippocampal sclerosis.

AMYLOID IMAGING: PLAQUES COME ALIVE

Beginning in the mid-1990s, it occurred to some investigators that amyloid stains could be radiolabeled and used, with PET imaging, to visualize amyloid plaques in living humans. Klunk and Mathis (285) struggled for more than 10 years with Congo Red derivatives, including its salicylic acid derivative, Chrysamine G, as well as chemically distinct analogs such as the styrylbenzene x-34, before eventually abandoning that direction in 1999, due to the failure of all to penetrate the blood-brain barrier sufficiently. They then turned to thioflavin T as a starting point and eventually synthesized and tested more than 350 such benzothiazole compounds before settling on 2, which eventually became known as Pittsburgh compound A and B, respectively the 7th and 23rd of their compounds. The challenge was to make a substance that strongly bound to amyloid and at the same time with enough lipophilicity to readily cross the lipid membranes in the blood vessel walls but not so hydrophobic that it would not dissolve in aqueous solution or not so lipophilic that it that it would not readily wash out of white matter. An interesting side note is that early in vitro binding studies had shown that Pittsburgh compound B (PiB) bound to Aβ in AD brain homogenates (a mixture of Aβ 40 and 42 as well as other forms) in a 1:1 ratio while binding to synthetic Aβ 40 fibrils was almost a full magnitude lower, at 1:370.

The first human tests were done in Sweden, then the world leader in the “microdosing” required for radioactive PET ligands. PiB was chosen over PiA as it had a more rapid brain clearance. In January 2002, Agneta Nordberg chose as the first human subject a relatively young woman whose memory problems had forced her to stop working as a healthcare professional, and in July 2002 the results on the first 14 subjects, including 9 with AD and 5 controls, were presented at the International Conference on Alzheimer’s Disease in Stockholm. A news story in Science described the impact of the finding, “At the conference, images of the results audibly took the audience’s breath away: In healthy people, the marker sailed right through the brain and. . . . in people with early Alzheimer’s disease, the marker stuck in the cortex….”.

Much of the subsequent work over the next several years focused on comparing scan findings in mentally normal elderly versus those with AD dementia. This was summarized in 2009 in a review by Wolk and Klunk (286). “Not all patients with clinically diagnosed AD have been reported to have levels of PiB binding that differ from controls.” “Presumably, these cases reflect misdiagnosis…” and “Several groups have now reported that 20-30% of healthy [older] controls have elevated PiB binding.” The relevance of PiB binding for AD was stated as follows. “The most important role of amyloid imaging …[will be]… facilitating the development of effective disease-modifying therapies” and “The finding that MCI patients often have levels of PiB retention similar to that seen in AD, coupled with the finding that at least 25% of cognitively normal elderly have measurable PiB retention, suggests that we will need to look in asymptomatic people to find the earliest stages of AD pathology. It may be that we have to identify and treat people at this early stage to achieve substantial disease-modifying effects.” They also noted that amyloid imaging could meet “…the need in clinical trials for more specific and less variable markers of progression than the currently-used clinical measures, which would allow for accurate assessment of efficacy in potentially smaller cohorts of patients,” especially as amyloid imaging could limit the selection of subjects to those with positive cortical amyloid scans.

Meanwhile, other amyloid ligands were under development, as well as a move to F18 radioactive labeling rather than the C11 isotope used with PiB. The 20-minute half-life of C11 limited its use to large medical centers with an on-site cyclotron, while F18’s 109-minute half-life meant that it could be synthesized and distributed by air transport over whole regions of the country. PiB was eventually labeled with F18 and licensed to GE, renamed as GE-067 and later flutemetamol. Others took a different approach, noting that stilbenes were structurally similar to benzothiazoles. Stilbenes were first reported as potential amyloid ligands in 2001 (287, 288). Of these, the most promising was SB-13, which was labeled with C11 and used in human brain imaging by 2005 (289, 290). Labeling SB-13 with F18 made it too lipophilic but similar compounds retained desirable characteristics after the label, including AV-1, developed by Avid Radiopharmaceuticals, later licensed to Bayer, renamed as BAY94-9172 and still later as florbetaben. Further work by Avid turned to yet another class of structural analogs, the styrylpyridines. One of these, AV-45, was successfully labeled with F18 and renamed florbetapir.

Continuing work with PiB, flutemetamol, florbetaben, and florbetapir found them generally to have similar binding affinities and results in human subjects and an overall picture of amyloid imaging in AD versus normal older people was being more completely fleshed out. Increased PET signals in the nondemented elderly were primarily in areas that also showed the heaviest binding in AD, all along the cortex of the medial surface of the interhemispheric fissure, but sparing the pre- and postcentral gyri, thus including the anterior and posterior cingulate gyri and their neighboring areas, respectively the medial frontal cortex and precuneus. This was somewhat puzzling to neuropathologists, as most of the work of the past several decades had identified the basal frontal and temporal cortex of the cerebral convexities as being the most and earliest affected, although this could perhaps have been an artifact of the more limited sampling done by neuropathologists compared to the whole-brain images available with PET. However, even Braak and Thal had found basal areas of the frontal and especially temporal lobe to be the earliest involved cortical regions. A possible exception was the work of Brun and Englund (134), who had described a primarily temporo-parietal convexity focus, although they were describing the net changes including not only plaques and tangles but also neuronal loss and gliosis. Otherwise, however, amyloid imaging was confirming neuropathologists’ work. While the imagers were initially surprised by the high prevalence of positive scans in the nondemented elderly, this was conforming to what neuropathologists had long known. Although reports with different agents and from different centers varied somewhat, most found that between 10% and 30% of normal older people had positive amyloid PET scans. As found in postmortem brains, prevalence increased with age. Mielke et al (291), in a 2012 study involving 483 cognitively normal subjects aged between 70 and 92, reported that 44% had increased cortical PiB uptake. In 2013, Mathis et al (292) reported a 55% prevalence in those over 80.

In sharp contrast to the hundreds of living subjects studied with amyloid PET, by 2008, there had been only 2 individuals with autopsy case reports (285). One of these (293) was a meticulous correlation of PiB antemortem imaging and postmortem histology and biochemistry for Aβ in 19 sites in the single brain examined, with highly significant correlation coefficients ranging from 0.73 to 0.86. Also performed were histological colocalization studies on adjacent sections, showing a one-to-one correspondence of plaques shown with either fluorescently-labeled PiB, thioflavin S or X-34. Additionally, occasional correspondence could also be seen within blood vessels with amyloid angiopathy, individual diffuse plaques and individual neurofibrillary tangles, although in quantitative comparisons there was no significant correlation of antemortem PiB signal with the amount of phosphorylated tau pathology by IHC. Earlier studies with fluorescent PiB by Klunk and Mathis had indicated binding in human brain sections to classical senile plaques and amyloidotic blood vessels (285) while in 2007 Lockhart et al (294) had reported, using concentrations of PiB similar to what would be encountered in the brain during imaging, a one-to-one correspondence of both diffuse and classical plaques (Fig. 6), as seen with H3 PiB autoradiography, Aβ IHC and thioflavin S. While the colocalization of PiB to neurofibrillary tangles was not statistically significant on quantitative comparison, the binding to diffuse plaques was very strong, equivalent to that seen in classical plaques. Diffuse plaques were unambiguously identified not only by adjacent sections stained with thioflavin S and Aβ IHC, but also by their characteristic location as a subpial “band” in the cortical molecular layer, as described by earlier investigators. Cerebellar diffuse plaques, however, were not identified by either group, with either fluorescent or radiolabeled PiB, again underlining their “very diffuse” nature. In a later study by the Lockhart group (295), another autoradiographic binding study showed equivalent results between H3 PiB and C14 SB-13 while H3 FDDNP showed no binding to either plaques or tangles. Again, there was unambiguous and strong binding to a characteristic site for diffuse plaques, this time in the parvocellular parasubicular cell islands, as described years earlier by Kalus et al (92) and Akiyama et al (93). Another single-case autopsy study (296) confirmed Ikonomovic’s work by correlating Aβ IHC with antemortem PiB retention in 17 sites, again finding this to be strong and significant (R = 0.899). Bielschowsky staining showed mostly sparse neuritic plaques while diffuse plaques were frequent, leading the investigators to state that “…PiB uptake on PET is a specific marker for Aβ density but cannot differentiate neuritic from diffuse amyloid plaques…”. In contrast, another single-case autopsy study by Cairns et al had shown almost an opposite finding, as a subject rated as negative for PiB died a year and a half later and was found to have frequent neocortical diffuse plaques but only sparse neuritic plaques (297). Other groups had achieved tissue correlations using cortical biopsy done for therapeutic reasons in patients with normal pressure hydrocephalus (NPH). A total of 52 patients with suspected NPH had both cortical biopsy and amyloid imaging with F18 flutemetamol (298). Correlations of antemortem ligand retention with postmortem plaque histology were strongly significant, with amyloid imaging showing a sensitivity and specificity of 93% and 100% respectively, for histologically demonstrated amyloid presence.

FIGURE 6.

Frontal cortex sections demonstrating concordance between H3 PiB autoradiography (D) and 6E10 monoclonal antibody IHC for Aβ (F), from Lockhart et al (2007). Arrows demonstrate specific plaques marked by both PiB autoradiography and Aβ immunostaining. NDB = section treated with the nonradioactive competitive ligand BTA-1. Reprinted from Ref. (294), copyright Oxford University Press.

In 2008, the US Food and Drug Administration jolted the amyloid imaging community by requiring autopsy proof of ligand accuracy as a condition for licensing. Avid Radiopharmaceuticals was the quickest out of the gate, publishing their Phase III findings (Fig. 7) on 35 subjects imaged with F18 florbetapir in January 2011 (299), with an extension to 59 subjects, published in June 2012 (300), receiving the historic first FDA approval of amyloid PET (marketed as Amyvid). Within the next 2 years, Bayer Healthcare/Piramal and GE Healthcare, both obtained FDA approval for F18 florbetaben (Neuraceq) and F18 flutemetamol (Vizamyl), respectively, after similar Phase III trials (301, 302). As the FDA had stipulated that the standard of proof for amyloid imaging agents would be autopsy confirmation of their sensitivity for the clinically relevant neuritic and cored plaques at or above clinically significant (CERAD moderate or frequent) densities (rather than diffuse plaques, with uncertain or weak clinical significance), the Bielschowsky stain was used by all 3 trials. As requested by the FDA, the amyloid scans were interpreted by readers as simply positive or negative, based on the appearance of the entire cerebral cortex. The sensitivity and specificity of positive scans was very high, for all 3 compounds, for finding CERAD moderate or frequent neuritic plaques (175) but none of the methods was sensitive enough to reliably identify sparse plaque densities (300–302). Despite this, amyloid PET has allowed a significant improvement in clinical AD diagnostic accuracy, estimated at over 80%, as compared with 70% for standard neurological examination alone (276, 277).

FIGURE 7.

Paired F18 Florbetapir PET amyloid scans and Aβ immunohistochemistry from the same subjects after death. Reprinted from Ref. (299).

As mentioned previously, the establishment of both staging and grading of cancer has resulted, through many clinical trials, in the development of treatment plans that are customized to provide the most optimal care at each stage and grade of cancer. Thal amyloid staging has shown that amyloid “spreads” from an initially cortically confined location to subcortical and then brainstem and cerebellar structures, in concert with progressive cognitive deterioration and higher Braak neurofibrillary stages (102, 303). As treatment of AD is likely to be more effective at early disease stages, it would be desirable, for clinical trials, to subdivide trial subjects into lower and higher pathology-based stages of AD as then trials could focus on the earlier-stage patients, or the data could be analyzed separately. An agent that failed to provide symptomatic or disease-modifying effects in higher stage AD patients might be found to be effective in lower stage AD patients. The amyloid progression through Thal phases potentially allows pathology-based clinical staging of AD using current PET amyloid imaging technology.

Although amyloid imaging generally does not reveal cerebellar plaques, with the possible exception of subjects with PS1 mutations (304), several PET studies had noticed the presence of striatal amyloid. Much attention was drawn to reports of striatal plaques at early disease stages in subjects with early-onset, autosomal dominant inheritance of AD (305–309), but a few groups had also documented striatal amyloid in late-onset sporadic AD (310, 311) and one of these found the striatal amyloid signal in late-onset subjects to be equivalent to that in early-onset disease (312). Studies of autopsied demented and nondemented elderly subjects (303) calculated that a positive striatal amyloid imaging signal, based on the assumption that only moderate and frequent striatal plaque densities would be detected, could be capable of predicting a higher Braak neurofibrillary stage as well as the presence of dementia and clinicopathological AD. However, doubts remained about how reliably amyloid imaging could pick up a positive striatal signal, due to the high likelihood of significant signal confounding by high partial volume effects in this relatively small structure.

To assess the reliability of a positive striatal amyloid imaging signal for the presence of striatal plaques, data from the Phase III FDA-monitored clinical trial of the GE compound, F18 flutemetamol (301), was used. This study compared the presence of a qualitatively positive flutemetamol PET striatal amyloid imaging signal (Fig. 8), as determined by the majority decision of 5 radiologists, with the presence of postmortem striatal amyloid deposits as determined by 2 neuropathologists (313). The sensitivity of flutemetamol PET striatal amyloid imaging for the presence of several defined density levels of histologically demonstrated striatal amyloid deposits ranged between 67% and 91% while the specificity ranged between 96% and 100%. The study concluded that a majority-positive striatal flutemetamol PET signal reliably predicted the presence of significant striatal amyloid deposits. A later analysis of the same study with additional subjects found that a combination of neocortical and striatal SUVr thresholds could distinguish Thal amyloid phases greater than 3, as defined by postmortem histology (314), thereby allowing, for the first time, pathology-based clinical AD staging. Other groups (315–318) have also reported promising data supporting PET amyloid-based Thal phase assignment. Due to the insufficient sensitivity of current PET amyloid tracers, however, the detection of Thal phases 1 and 2 is possible for less than half of such subjects.

FIGURE 8.

Negative (A) compared to highly positive (B) striatal F18 flutemetamol PET signal, in the coronal plane, with MR images for orientation. The striatal signal is centered over the gray matter of the head of the caudate nucleus and putamen (arrows). Reprinted from Ref. (313), with permission from IOS Press and the Journal of Alzheimer’s Disease.

LIMITATIONS OF AMYLOID IMAGING

The accumulated data of the autopsy-validated clinical trials for florbetapir, florbetaben, and flutemetamol (300–302, 314–318), suggest that the sensitivity of amyloid imaging is not yet great enough to reliably detect the earliest stage of AD, which might be considered to be CERAD sparse cortical diffuse and/or neuritic plaque densities when plaques are limited to the cerebral cortex and mesial temporal lobe. This has implications for AD prevention studies as it is widely assumed that it will be easier to stop the disease at a time when plaques are still sparse and not widespread. Lowering the SUVr thresholds for amyloid presence might increase the sensitivity for cases with sparse amyloid but would be likely to commensurately decrease specificity. There is therefore an urgent need for more sensitive PET amyloid tracers.

Another concern for amyloid imaging has been the revised NIA criteria for assessing whether AD neuropathological changes can be considered the cause, in any single individual, of either MCI and AD (212, 284). These criteria include Thal amyloid staging in the tripartite classification, along with Braak neurofibrillary stage and CERAD neuritic plaque density. As Thal staging does not distinguish between diffuse and classical plaques, this opens the possibility for a subject with only diffuse plaques to be diagnosed as AD-MCI or AD-dementia. It has been commonly assumed by the amyloid imaging community that because amyloid imaging cannot detect diffuse plaques (because they were assumed not to contain “fibrillar” amyloid), such cases would not be diagnosed. This seems unlikely however, as much previous work (as mentioned in preceding sections), indicates that diffuse plaques actually do contain ultrastructurally identified amyloid fibrils (although relatively sparse amounts) and stain positively with the β-pleated sheet-specific thioflavin S (except for cerebellar diffuse plaques, which have only rare amyloid fibers on EM and stain inconsistently with thioflavin S; see previous sections). A close reexamination of the postmortem-validated Phase III flutemetamol trial (319) as well as recent reports by others (318) have confirmed that diffuse plaques do affect the PET amyloid neocortical SUVr in proportion to their density and distribution. It is likely, however, that at least in the neocortex, whenever moderate or frequent diffuse plaques are present there will be comparable densities of neuritic and cored plaques and so it seems that the envisioned diagnostic conundrum will be infrequent. To test how often such a scenario might exist, NACC data from a prior postmortem study were examined (276). Of 707 cases with moderate or frequent cortical diffuse plaques, only 76 (11%) had less-than-frequent cortical CERAD neuritic plaque densities.

Another potential shortcoming is that PET amyloid tracers appear to bind preferentially to Aβ42 but not Aβ40, leaving open the possibility that PET amyloid might underestimate the amyloid load in subjects with predominantly Aβ40-containing plaques. As mentioned, Klunk and Mathis had found an almost 400-fold difference in the in vitro binding affinity of PiB with Aβ40 versus Aβ42 (285). ELISA analysis of frozen frontal pole samples showed a significant correlation with florbetapir PET SUVr for Aβ42, but not Aβ40 (320). As the concentrations of Aβ42 in both aging and AD brain tissue are most often several-fold higher than those of Aβ40 (194, 321, 322), this is not likely to be a common problem for PET amyloid imaging but there are reports that patients treated with the immunotherapy agent, bapineuzumab, or the γ-secretase inhibitor, semagacestat, have an increased cortical Aβ40/Aβ42 ratio (323, 324). Further studies are warranted to determine whether this may be a frequent finding after anti-amyloid therapies, and if so, whether it might be of a magnitude that might affect PET amyloid imaging quantification.

A CHANGING ROLE FOR PLAQUES IN CONCEPTS OF AGING AND AD

Although tangles have been recognized as having a greater contribution to the severity of dementia, this may be mostly due to their later occurrence. When both demented and nondemented subjects are included, plaques are strongly associated with cognitive test scores, as first shown almost 60 years ago by Sir Martin Roth and colleagues (46, 47). Etiologically, the amyloid hypothesis of AD has remained dominant, stipulating that plaques are the first and central event (325), with tangles and neurodegeneration following. The postmortem neuropathology of aging and sporadic AD, however, has clearly shown that mesial temporal lobe neurofibrillary tangles are universal in aging humans by age 70 or earlier (166) while plaques, on average, never quite reach universality (161, 162). Furthermore, there is an interdependence of the 2 lesions, as in sporadic AD, widespread neocortical tangles never occur unless neocortical plaques are also widespread. And as so elegantly elucidated by Heiko and Eva Braak, the stereotyped and intricate neuroanatomical pattern formed by mesial temporal lobe tangles does not differ between normal aging and sporadic AD, making it difficult to argue that tangles are not the first and necessary step. The recent christening of inevitable, age-related mesial temporal lobe tangles as “primary age-related tauopathy,” or “PART” has reignited these long-simmering issues (326–329).

It is impossible to deny the primacy of plaques in autosomal dominant AD due to presenilin or β-APP mutations, but it also seems possible that sporadic AD may have a different origin, with tangle formation in selected neuronal subdivisions leading to a chain of molecular events culminating in neocortical plaque formation. Nucleus basalis of Meynert tangles, for example, occur contemporaneously with entorhinal or even transentorhinal tangles (330, 331) and the subsequent loss of their cholinergic input could lead to dysregulation of β-APP processing in postsynaptic neocortical neurons (332, 333). It is known that one type of aggregated protein may “cross-seed” the aggregation of a different one, so one might speculate that, as suggested by Pearson et al (135) in their “neuroanatomical spread hypothesis,” that “…the disease process may extend along the connecting fibers.” Arendt et al (139) even suggested that neocortical neuritic plaques might represent the degenerating synaptic terminals of tangle-bearing neurons of the nucleus basalis of Meynert. These investigators thus presaged the “prion hypothesis” of neurodegenerative disease transmission that has undergone a dramatic revival in recent years. This has raised the suggestion that the aggregated proteins present in neurodegenerative disease might be transmissible between humans. A recent committee of expert neuropathologists concluded, however, that “there is currently insufficient evidence to suggest more than a negligible risk, if any, of a direct infectious etiology for the human neurodegenerative disorders…” but emphasizing the need of “… further investigation” (334).

There must be an explanation, however, as to why plaques, unlike tangles, do not form in every single aged human. This may conceivably be due to individual genetic backgrounds that are permissive or nonpermissive for plaque formation. It has been known for almost 3 decades that the apolipoprotein E-ε4 allele increases the likelihood that plaques will form while the ε2 allele decreases the likelihood. This effect is dramatic, and could be the major factor determining what path individual humans follow. All humans develop tangles, but most of those with an ε4 allele go on to develop abundant plaques while most of those with an ε2 allele do not. Whatever the resolution may ultimately be, the inscrutable linkage of these 2 enigmatic lesions means, more than a century after Alzheimer, that both are necessary for the full development of the clinicopathological syndrome that bears his name, and both will remain major targets for prevention and therapy.

 

Thomas G. Beach and the Civin Laboratory for Neuropathology have received support from the National Institute on Aging (P30 AG19610), National Institute of Neurological Disorders and Stroke (U24 NS072026), the Arizona Department of Health Services, the Arizona Biomedical Research Commission, the Sun Health Foundation and the Michael J. Fox Foundation for Parkinson’s Research.

The authors have no duality or conflicts of interest to declare.