Classification and basic pathology of Alzheimer disease
- First Online:
- Cite this article as:
- Duyckaerts, C., Delatour, B. & Potier, M. Acta Neuropathol (2009) 118: 5. doi:10.1007/s00401-009-0532-1
- 2.3k Views
The lesions of Alzheimer disease include accumulation of proteins, losses of neurons and synapses, and alterations related to reactive processes. Extracellular Aβ accumulation occurs in the parenchyma as diffuse, focal or stellate deposits. It may involve the vessel walls of arteries, veins and capillaries. The cases in which the capillary vessel walls are affected have a higher probability of having one or two apoε 4 alleles. Parenchymal as well as vascular Aβ deposition follows a stepwise progression. Tau accumulation, probably the best histopathological correlate of the clinical symptoms, takes three aspects: in the cell body of the neuron as neurofibrillary tangle, in the dendrites as neuropil threads, and in the axons forming the senile plaque neuritic corona. The progression of tau pathology is stepwise and stereotyped from the entorhinal cortex, through the hippocampus, to the isocortex. The neuronal loss is heterogeneous and area-specific. Its mechanism is still discussed. The timing of the synaptic loss, probably linked to Aβ peptide itself, maybe as oligomers, is also controversial. Various clinico-pathological types of Alzheimer disease have been described, according to the type of the lesions (plaque only and tangle predominant), the type of onset (focal onset), the cause (genetic or sporadic) and the associated lesions (Lewy bodies, vascular lesions, hippocampal sclerosis, TDP-43 inclusions and argyrophilic grain disease).
The pathology of Alzheimer disease (AD) has been studied intensely in the last 20 years. Animal models have provided valuable information to understand the pathogenetic mechanisms. It has become impossible to review this abundant literature in a single paper. This is the reason why we will not systematically consider the animal models (the reader is referred to another review by the same authors ).
Alzheimer disease pathology may be divided into three broad chapters: lesions related to accumulation (“positive lesions”), those that are due to losses (“negative lesions”) and finally those that are due to the reactive processes (inflammation and plasticity). The first ones are robust, easy to detect and constitute the basis of the diagnosis. The losses—of neurons and of synapses—are difficult to evaluate; they do not belong to the diagnostic criteria but could be the alterations that are more directly related to the cognitive deficit. Since macroscopic findings are due to losses, they will be considered after the microscopic data.
Lesions by accumulation
The lesions related to accumulation are mainly constituted by the extracellular deposition of Aβ peptide and the intracellular aggregation of tau protein. Both Aβ peptides and tau protein are normal cellular constituents, both take, in AD, an abnormal fibrillary structure, associated with a poor solubility in water solution. However, the relationship between the two types of lesions remains elusive.
Aβ peptide deposition
Aβ production and clearance
The deposition of Aβ is related to an imbalance between its production and clearance. Aβ peptide is cleaved from its precursor, the amyloid protein precursor or APP, a transmembrane protein, by two sequential enzymatic activities: β-secretase that cleaves APP in its extracellular domain and leaves a fragment called C99 which is secondarily cleaved by the transmembrane gamma-secretase complex that includes presenilin . The subcellular compartmentalization of these successive cleavages remains discussed. It is, today, generally accepted that the last steps take place in the endosomal–lysosomal pathway. Actually the deposits contain a mixture of various Aβ isoforms, some starting at amino-acid (AA) 11 (“N-truncated” isoforms)  or ending at AA 39 to 42. A common form starts at position 3 with the initial glutamine residue having been cyclized into pyroglutamate . The isoform that ends at AA 42 contains two supplementary amino-acids at the C-terminus of the peptide, i.e. contains a larger segment embedded in the cell membrane; it is more hydrophobic and prone to precipitate in water solution. Mutations of APP, presenilins 1 and 2 have been associated with an increase in Aβ production and in the ratio of Aβ42/Aβ40.
The function of Aβ is unknown; the peptide that is physiologically produced at low concentration has been considered a waste product. It has recently been found to act positively on synapses at picomolar concentration . A ligand of Aβ has not been firmly identified yet but it has been shown that Aβ oligomers act through glutamate receptors: oligomers have been isolated from AD brains and shown to rapidly inhibit long-term potentiation (LTP), enhance long-term depression (LTD) and reduce dendritic spine density. This effect was attributed to dimers. NMDA receptors were necessary for the change in spine density while the metabotropic glutamate receptors were required for the LTD enhancement .
The Aβ peptide can be degraded by multiple enzymes among which neprilysin and insulin degrading enzyme (IDE) are probably the most important. It is, on the other hand, cleared from the brain parenchyma through the interstitial fluid and Virchow–Robin perivascular space, a space that is considered to play the role of a lymphatic vessel (for review see ). The peptide interacts with the low-density lipoprotein receptor-related protein (LRP)-1  or with P-glycoprotein  to cross the blood–brain barrier. There is also a flux of Aβ peptide from the blood to the cerebral parenchyma mediated by the receptor for advanced glycation endproducts (RAGE) .
Types of deposits
Many anti-Aβ antibodies have been developed. Used as immunohistochemical reagents, they have revealed different aspects of the pathology. Some of them interact with the last amino-acids located at the C-terminus of the peptide and allow distinguishing the Aβ40 and Aβ42 isoforms, other ones do not react with the N-truncated isoforms.
The parenchymal deposits
The parenchymal Aβ deposits are associated with various proteins, lipids and cells: ApoE is an early and common component of the various types of Aβ deposits [244, 325] (Fig. 3f). ApoE is mainly produced by the astrocytes in the brain and is involved in cholesterol transport. Since the Apoε4 allele is the main risk factor of AD, the hypothesis that cholesterol was associated with Aβ peptide seemed plausible. Two studies, using cholesterol oxidase method and filipin staining, concluded that cholesterol is accumulated in the senile plaque [57, 236]; however, these results may be artifactual . ApoJ, also called clusterin, has been observed in focal deposits with a neuritic corona . Although APP is known to be a metal-binding protein, only few studies analyzed the content of the senile plaques in metal ions. Zinc and copper ions were found with Raman microscopy ; the presence of zinc was confirmed by immersion autometallographic detection using sodium sulfide . Iron has also been found in the core of the senile plaques . Various components of the extracellular matrix have been shown to accumulate in the senile plaque such as ICAM1 , thrombospondin , heparan sulfate proteoglycan . Alpha1-antichymotrypsin, a serine protease inhibitor synthesized in the liver, has been thoroughly studied. It consistently colocalizes with Aβ amyloid deposits  and could be involved in tau hyperphosphorylation . Cathepsin D, a lysosomal enzyme, is also abundant in the processes around the plaque and in the plaques; it could be related to the activation of the endosomal–lysosomal pathway .
Diffuse Aβ deposits Diffuse Aβ deposits are poorly immunoreactive so that their apparent number depends on the quality of the immunohistochemistry (Fig. 2a). They are usually large (from 50 μm to several hundred μm) and ill-limited. In some regions of the brains, the deposits are only of the diffuse type. In the presubiculum, the Aβ deposit forms a continuous and large area of immunoreactivity (IR), named “lake-like” ; in the internal layers of the entorhinal cortex the large diffuse deposits are called “fleecy” . In the striatum and molecular layer of the cerebellum, the deposits are generally of the diffuse type. They may also be considered as “diffuse” in the subpial region of the isocortex where they may form an uninterrupted band of immunopositivity. Diffuse deposits have been found in large numbers in subjects whose intellectual status had been evaluated as normal—leading to the conclusion that these lesions may not be directly toxic [82, 88]. The status of these cases of “pathological aging” remains uncertain. It could be the premise of a full-blown AD but it is not known how long diffuse deposits may remain uncomplicated in the brain. They could explain why Pittsburgh compound B (PIB) fixation is currently found in vivo in persons who are apparently free from any significant deficit. Diffuse deposits indeed bind PIB although not specifically . Diffuse deposits are associated with Apolipoprotein E (ApoE) . The negativity of some diffuse deposits to antibodies directed against the N-terminal part of the molecule has been considered the consequence of ApoE binding with the N-terminal part of Aβ peptide in the early stage of deposition .
Focal deposits In addition to diffuse deposits, immunohistochemistry shows dense and spherical accumulations of Aβ peptide, the focal deposits. The focal deposits are visible in slides stained with hematein and eosin (Fig. 3a) while diffuse deposits are unstained. Focal deposits may be amyloid or not, associated or not to a neuritic corona.
Amyloid or not Some focal deposits do not have the amyloid tinctorial characteristics but are easily revealed by Aβ immunohistochemistry. They are said to mainly contain the Aβ-40 peptide . Other focal deposits are amyloid, revealed by Congo red or thioflavin S. CD-68 positive, activated microglia nearly always accompany the amyloid focal deposits (Fig. 3b, e) . This observation could be explained by two different scenarios: either the microglial cell is itself necessary to the transformation of the diffuse Aβ deposit into amyloid  or it is the amyloid conformation that activates the microglia. Numerous microglial receptors (which are outside the scope of this review) have been shown to be involved in this response. Microglial cells secrete cytokines that trigger and sustain a chronic non-immune inflammation (see below “Inflammation: relation with immunotherapy”).
Amyloid deposits with a neuritic corona The focal deposits that constitute the core of the neuritic plaques are virtually always amyloid (Fig. 3c, e). The core is surrounded by a clear halo that separates it from a light and diffuse zone of IR constituting the outermost ring of the deposit. These “cored deposits” (Fig. 3b) are said to be predominantly made of Aβ-42 . The quantity of Aβ accumulated by deposit has recently been assessed by mass spectrometry: a microdissected plaque disc has been said to contain 50–100 fmole of Aβ peptide , a quantity that represents roughly 3 to 6.1010 molecules of Aβ i.e. 30 to 60 billions. It may be wondered how such a large quantity of molecules can assemble in such a short time: it has indeed been shown in APP transgenic mice that the plaque takes only a few hours to develop . The presence of one, or more commonly of several microglial cells, is directly associated with the amyloid core: they are usually located in the clear halo that separates the core from the outermost ring of diffuse IR that constitutes the external halo of the focal deposit (Fig. 3e). The corona that surrounds the focal Aβ deposit contains neuritic and astrocytic components. The Aβ focal deposit together with the neuritic corona constitutes the “mature”, “classical” or “neuritic plaque”. When is the Aβ core of the senile plaque being surrounded by neurites? In a study analyzing the various forms of deposits in a population of prospectively assessed cases, it was found that neuritic plaques were observed only in those samples in which congophilic deposits were also present . It has long been known that deposits were generally devoid of a neuritic corona in areas where tangles were absent .
The neuritic component of the corona The presence of neuronal processes surrounding the plaque core has been recognized in the early days of neuropathology by Redlich , Fisher  and Cajal . Bielschowsky technique was initially used but other methods such as Bodian’s and, with better selectivity, Gallyas’, were also able to show the processes. The nature of the processes—axonal or dendritic—was however difficult to elucidate with these methods, all the more as Golgi techniques, which enable to visualize the whole dendritic tree and sometimes the axon of the neuron, were not efficient to visualize the plaques. It was however apparent with the silver methods that the neurites surrounding the plaque core were abnormally large, so-called “dystrophic”, leading Cajal to conclude that they were actually regenerating processes . Ultrastructurally, the so-called degenerating neurites that make the corona of the plaques are distended; they contain large amounts of lipofuscin and dense bodies, degenerating mitochondria [209, 309], and paired helical filaments (PHF)  made of tau protein (see later). The presence of synaptic vesicles  suggests the development of aberrant connections and demonstrates that the senile plaque is not an amorphous structure but a living lesion. Immunohistochemical and electron microscopy evidences suggest that most of the neurites are axonal: they are indeed labeled by neurofilaments but not by MAP2 antibodies  and contain synaptic vesicles. The origin of these axons is only partially known in man, except for special cases: there are, for instance, good evidences that the dystrophic axons of the plaques seen in the superficial part of the molecular layer of the dentate gyrus originate in the superficial layer of the entorhinal cortex [99, 166]. In APP transgenic mice, the tracking of the connections that innervate the senile plaques shows that most of them are cortico-cortical [83, 290]. In another transgenic model, entorhinal axons form dystrophic boutons in contact with Aβ deposits located in the molecular layer of the dentate gyrus .
Tau antibodies label the processes of the corona. The intensity of the labeling and the number of processes labeled depend on the sensitivity of the antibody, AT8 being one of the most sensitive ones. Only a subset of the neurites is tau positive (the type 1 dystrophic neurites according to Wang and Munoz  (Fig. 3c)). The APP-positive processes (Fig. 3d) are large, globoid, and less numerous than the tau-positive ones; they are also chromogranin positive (type 2 dystrophic neurites ). Only a minority of neurites are tau and chromogranin immunoreactive . Various hypotheses have been made to explain the diversity of these neurites. Thal et al.  consider that the accumulation of APP in the neurites is the consequence of an alteration in axonal transport related to tau pathology, while Wang and Munoz “propose a hierarchical model of plaque formation in which Aβ deposits do not incorporate tau neurites unless neurites bearing synaptic proteins and beta APP are also present” .
The C-terminal domain of APP, phosphorylated on the threonin 668 is associated with tau IR and could be the missing link between tau and Aβ pathology . Neurites may also contain ubiquitin (sometimes in large globular processes, probably at an early stage of neuritic alteration) , or neurofilaments . Dystrophic ChAT positive, sprouting processes around the core of the senile plaque may be numerous; their number is correlated with the Aβ 1-42 levels measured in the same cortical samples .
The astrocytic component Neuritic plaques are surrounded by astrocytic processes; contrarily to microglial cells that are in contact with the amyloid core, the fibrous astrocytes are located at the periphery of the plaque. There are indications that the astrocytic component appears late in the history of the senile plaque; the abundance of astrocytic processes penetrating the core of the plaque could possibly be related to a regressive stage of the lesion (‘‘remnant plaque”) .
Special type of deposits “Cotton wool plaques” are focal deposits that are easily identified after hematein–eosin stain as round, homogeneous and eosinophilic structures. They are made of Aβ-42 and contain only sparse glial components. They are not congophilic and are not surrounded by a corona of neurites. Neurites may however penetrate them and constitute a ball of interweaving processes. They have initially been observed in familial AD due to deletion of the exon 9 of the presenilin 1 (PS1) gene [70, 213] but have been seen (although rarely) in sporadic AD cases, always associated with typical plaques . The main species that accumulates in cotton wool plaque is N-truncated . Another special type of plaques, the “inflammatory plaque”, characterized by a strong microglial reaction, surrounding an amyloid core without Aβ reactivity, has also been seen in PS1 mutations .
Differential distribution of the deposits with and without neuritic corona The aspect of a given deposit depends on its stage and topography. The morphological changes of the deposits over time are not directly accessible; they are reconstructed from the observation of different types of lesions in cases or in regions at various stages of severity. It has been hypothesized that deposits, in the isocortex, were initially diffuse, then focal and Congo red positive, finally surrounded by the neuritic corona that defines, with the amyloid core, the mature (neuritic) senile plaque . Amyloid deposits, diffuse, focal and neuritic are also found in the olfactory bulb at an early stage of the disease . There are however topography in which neuritic senile plaques are nearly never seen: the presubiculum where large areas of immuno-reactivity constitute the lake-like deposits , the striatum or the molecular layer of the cerebellum for instance. The fleecy deposits found in the deep layers of the entorhinal cortex are made of N-truncated Aβ-42 peptide  and may be phagocytosed by astrocytes (Fig. 2c) . Aβ peptide IR has been found in the skin  but with no relationship with AD . As already mentioned, tracking experiments of the neurites surrounding the plaques, in APP transgenic mice, suggest that only some populations of axons may take part to the corona of the plaque . Cortico-cortical connections for instance may participate while thalamo-cortical connections are apparently never involved. This implies that component(s) of the amyloid deposit constitute(s) a message that is able to attract, to immobilize or to alter only some types of axons.
The vascular deposits
Sometimes the accumulation of Aβ peptide takes place selectively in the vessel walls; in such cases, lobar hemorrhages and small cortical infarcts may constitute the main pathology. This is, for instance, the case in the hereditary cerebral hemorrhages Dutch type, related to a mutation of the APP gene  leading to a change in AA 21 of Aβ. Other mutations (Flemish and Iowa) are also associated with severe cerebral amyloid angiopathy (CAA); they involve the same region of the APP gene. More recently severe CAA has been linked to the microduplication of the APP gene . Aβ peptide is first deposited around the basement membrane and then in the media. In the most advanced cases, the vessel takes the aspect of a double barrel and fibrinoid necrosis may be found in the media, this last alteration being more frequently observed in cases with hemorrhages . The prevalence of cerebral amyloid angiopathy (CAA) has been diversely appreciated. In a large autopsy series (n = 113; 61.1% female, 55.8% clinically demented, age range 54–102 years, mean ± SE 83.5 ±0.93 years), Attems et al.  found that 24% of the cases with AD pathology showed no CAA, while 23.5% of the cases without AD pathology had CAA; Joachim et al.  considered that “at least a minimal degree of amyloid angiopathy was found in every brain showing histopathological abnormalities of AD”. When all intracerebral hemorrhages are considered, there is no increased risk in the group with CAA (since the weight of other risk factors, particularly hypertension, is overwhelming). Intracerebral hemorrhages in the cases with CAA are usually lobar, while those associated with hypertension involve the basal ganglia .
The vascular deposition predominantly made of Aβ-40  takes place in arteries and capillaries (see later) but the veins are also involved (for review see ). The proportion of vessels with Aβ deposit and their topography are quite variable, the occipital cortex being most massively and frequently involved ; the occipital involvement is correlated with AD pathology .
Angiopathy without capillary involvement (type 2 CAA of Thal et al. ) This form is the most frequent. There is no relationship of type 2 CAA neither with ApoE genotype  nor with AD pathology .
Angiopathy with capillary involvement (type 1 CAA of Thal et al.) The involvement of the capillary walls is observed in only some cases of amyloid angiopathy, defined as type 1 cerebral amyloid angiopathy . This aspect where the amyloid seems to get out of the vessels lead the first observers to consider that there was an alteration of the blood–brain barrier (barrier = horos) hence the term dyshoric angiopathy. Capillary involvement is linked with the presence of one or two ApoE4 alleles (4 times more frequent than in type 2 CAA) . The use of antibodies specifically directed against Aβ 40 and Aβ 42 refines this distinction: the capillary CAA (capCAA) may be made of Aβ42 and, in this case, involves the perivascular space and the glia limitans. This pathological type is well correlated with AD pathology. Aβ 40, by contrast, is confined to the vessel wall itself. Aβ40 capCAA is not correlated with AD pathology.
The question of intracellular Aβ
The presence of Aβ peptide in the neuronal cell body (iAβ) has been much debated. There is currently considerable controversy concerning the very existence and, if admitted, the meaning of iAβ. In a paper in preparation, the team of Irina Alafuzoff (Leena et al., in preparation) has reviewed a large amount of the literature on this topic. We are grateful to the authors to have let us read their thorough analysis.
The conclusion that Aβ accumulates in the cell body of the neuron is essentially drawn from immunohistochemical experiments. The quality of the labeling depends on many variables among which the nature of the pre-treatment applied to retrieve the antigenicity. The use of heat pre-treatment enhances the visualization of iAβ, but formic acid impedes it [72, 73, 252]. Formic acid is commonly used to enhance the IR of extracellular Aβ. The antibodies directed against the Aβ peptide may either target an epitope located within the Aβ sequence: in this case, at least in theory, they will recognize, not only the Aβ peptide, but also full-length APP and all the APP fragments in which the epitope sequence is included. Antibodies raised against the first AA with the initial NH2 may in theory recognize both the Aβ peptide and the C-terminal fragment of APP after the β-cleavage. The antibodies raised against the last AA with the terminal COOH may only recognize either Aβ (40 or 42 according to its specificity) or truncated (α-cleaved) Aβ, also called P3 fragment.
Gouras et al.  used three sets of polyclonal antibodies selectively directed against the C-terminus of the Aβ40 and Aβ42 peptide  as well as the 4G8 monoclonal antibody that recognizes AA18-22 of the Aβ sequence. The results were similar. Intracellular granules were found with the Aβ42 and 4G8 antibodies. The labeling was stronger in AD cases than in age-matched or young controls and was said to be present in vulnerable regions before the development of full-blown pathology . The Aβ40 antibodies gave but a weak signal. Besides Gouras et al., several studies mention a positive intracellular signal in normal [306, 339] and AD cases [141, 191, 306, 339] with the 4G8 antibody. Such positivity is not sufficient to prove that it is the Aβ peptide that is present inside the cell. The recognized epitope could also be located in APP, C99 or C88 (the APP c-terminus after cleavage by the γ- and α-secretase, respectively), or in N-truncated Aβ. It has been repeatedly shown that there is no intracellular signal with antibodies that recognize the N-terminus of the Aβ peptide (such as the 6E10 monoclonal whose epitope is AA 4–9 and the 6F/3D antibody whose epitope is AA 10–15). Since iAβ is revealed by 4G8 and polyclonal antibodies against Aβ42, Wegiel et al.  conclude that the intracellular signal is due to the cleavage product of APP by α- and γ-secretase, made up of the AA 17–42 of the Aβ peptide sequence i.e. the P3 fragments. iAβ 42 IR is regularly found in Down syndrome patients. In the series of Gyure et al., intracellular immunolabeling was found with Aβ1–28 and Aβ40 antibodies even in young cases. Aβ42 IR was only present in the cases above 44 years of age, sometimes before the appearance of senile plaques, while there was no labeling in 75% of the controls . In the series of Mori et al., Aβ40 antibodies did not detect significant IR but all Aβ42 antibodies showed strong intraneuronal Aβ42 IR, especially in the youngest cases (e.g., 3 or 4 years old). Wegiel et al.  confirmed an early and strong IR with 4G8 and Aβ42 antibodies in trisomy 21 but did not find any relationship with the disease. Various pools of Aβ peptide can be isolated with different methods of Aβ peptide extraction from the tissue. Steinerman et al.  recognized four pools: extracellular and soluble (Tris buffer without detergent), intracellular and soluble (Triton X), membrane (Sodium dodecyl sulfate detergent, SDS), extracellular and insoluble (formic acid). Aβ40 and 42 were assayed using ELISA with an anti-Aβ11–28 capture antibody and with either Aβ40 or 42 detector antibodies in these four types of extracts. It was found that the clinical status was correlated with the Triton Aβ42 (soluble, intracellular) and the SDS Aβ42 (membrane) fractions. These data suggest that there is indeed some intracellular Aβ42 detectable in the neuron and that they are better correlated with the clinical status than the extracellular aggregates. However, it should be stressed that the method does not distinguish intraneuronal from intra-astrocytic Aβ. Aβ IR has indeed been found in the cell body of the astrocytes in close contact with Aβ diffuse deposits . Microdissection of neurons of the CA1 sector, followed by Aβ extraction in formic acid and ELISA of Aβ42 and Aβ40, showed a significant (although small) increase in Aβ42 (but not in Aβ40) in the AD group (especially in the 2 familial cases of the series) .
The accumulation of iAβ has been observed as granular deposits in old monkeys with 4G8 and several Aβ40, 42 and 43 antibodies  and in old dogs as a membrane labeling examined at electron microscopy using one polyclonal antibody recognizing AA 1 through 42 and a monoclonal antibody (G211) . In the transgenic models, Aβ intraneuronal accumulation is easy to identify and much simpler to distinguish from lipofuscin than in man. Large granules containing Aβ peptide IR have been seen within the cortical neurons of several transgenic mouse lines among which Tg2576 [133, 306], APPSLPS1M146L [193, 346], APPSLPS1 M146LKI , 3xTg-AD [250, 251] and 5XFAD . Whether the intraneuronal accumulation of Aβ peptide seen in transgenic mice is dependent on specific mutations of the transgene(s) or is the mere consequence of the overproduction of Aβ remains to be determined. The regularity with which Aβ peptide is found intracellularly in transgenic mice pleads, in our view, for the second possibility.
In triple transgenic mice, the removal of extracellular Aβ deposits (by immunotherapy) is shortly followed by the clearance of iAβ, indicating that there is a dynamic balance between the two pools . In transgenic animals, the density of iAβ decreases while the density of extracellular Aβ deposits increases [193, 251, 346] suggesting that the secretion of iAβ is responsible for its extracellular accumulation. Taking benefit of this observation, Christensen et al. have studied the neuronal loss in the frontal cortex in correlation with Aβ accumulation either in the neurons or in the extracellular space. They concluded that the neuronal loss was related to iAβ . If these observations were extrapolated to humans, iAβ would only be visible at an early stage, i.e. at a stage which cannot be explored by postmortem studies, except in exceptional cases. Accumulation of intracellular Aβ, in a multimeric form, has been observed in human fetal mixed brain cell cultures. Increasing the amount of exogenous Aβ peptide in the culture medium was associated with a decrease of multimeric beta-amyloid in the medium and its increase in the cells .
In conclusion, intraneuronal IR against the 42 C-terminal part of the Aβ peptide has been repeatedly mentioned in the literature. That this IR is evidence of the presence of full-sized Aβ in the cell and that it has some relationship with the disease remains to be firmly established in the human. Accumulation of Aβ42 in the neurons of transgenic animals appears to be better documented. However, this relationship could be caused by the strong overexpression of Aβ peptide in these animals, easily extrapolated to familial AD but maybe not applicable to sporadic AD.
Topography of the deposits
The large majority of the Aβ deposits are located in the gray matter although streaks of diffuse deposits may be seen in the white matter. The neuritic plaques are preferentially found in layers II and III . The areal topography of Aβ deposition depends on the stage of the disease and is not random. At least two ways of staging this progression have been published: in Braak scheme, only the cerebral cortex is taken into account: at stage A, amyloid deposits are found in the “basal portions” of the cortex; at stage B, all the isocortex is involved except the primary cortices; the hippocampus is mildly affected; at stage C, deposits can be seen in all areas of the isocortex including sensory and motor core fields . For Thal et al. there are five “phases” in parenchymal amyloid deposition: in summary, the isocortex is involved in phase 1, the hippocampus and the entorhinal cortex in phase 2, striatum and diencephalic nuclei in phase 3, various brainstem nuclei in phase 4, and finally the cerebellum as well as additional brainstem nuclei (pontine nuclei, locus coeruleus, parabrachial nuclei, reticulo-tegmental nucleus, dorsal tegmental nucleus, and oral and central raphe nuclei) in phase 5 . Thal et al. have identified, on the other hand, three stages in CAA, which do not correspond to the five phases of Aβ parenchyma that was just described. The vessels are affected in the isocortex at stage 1, in the allocortex, cerebellum and midbrain at stage 2, in the basal ganglia, thalamus, pons and medulla oblongata in stage 3 [313, 315]. Vessels of the cerebellum are involved earlier (vascular stage 2) than its parenchyma (phase 5). The vessels of the basal ganglia, thalamus, pons and medulla oblongata are affected later in the course (vascular stage 3).
The neurofibrillary tangles were first described by Alzheimer who discovered the accumulation of argyrophilic material in the cell body of the neurons after Bielschowsky silver stain. The term neurofibrillary stems from neurofibrils—a normal constituent of the neurons, also revealed by silver methods. The neurofibrils are made of neurofilaments and, indeed, the neurofilaments were among the first proteins to be looked for in neurofibrillary tangles (NFTs) . The tau revolution was preceded by the discovery that the normal constituent of the neuron, present in the NFTs, were neither the neurofilament proteins nor the tubulins . It turned out that the anti-tau antibodies were those that regularly labeled the NFT . Tau is a phosphoprotein containing a tubule binding domain comprised of three or four repeat regions (tau 3 R and 4 R) depending on the splicing of the RNA. The 3 and 4 R tau isoforms are found in AD.
The immunohistochemical finding of tau accumulation in AD neurons was rapidly followed by the biochemical confirmation of abnormally phosphorylated tau in isolated NFTs or homogenized samples from AD brains [80, 142, 143]: the number of phosphorylated sites is increased, some of them (such as Thr231 and Ser262) possibly specific (or at least suggestive ) to abnormal phosphorylation . Phosphorylated tau is actually detectable in normal brain parenchyma when the interval between biopsy and fixation is short  but aggregated tau is obviously absent. Tau is N-truncated in the process of tangle formation . The proteolysis of tau could be as important as its phosphorylation for the formation of tangles .
The accumulation of tau in the neuron is associated with the presence of cell cycle markers, such as cyclin B [237, 242] or cyclin dependent kinase 5 (CDK5) . They are related to “re-entry” in the cell cycle that could finally kill the neuron. In transgenic mice overexpressing normal human tau, there are also signs of re-entry: tau accumulation could thus be the cause of the re-entry rather than its consequence . On the other hand, in the xenopus oocyte, tau is hyperphosphorylated during mitosis . These observations suggest that tau and cell cycle are linked and that re-entry is implicated in neuronal death.
The accumulation of tau in the neuron has probably severe consequences on neuronal functions. Tangle-bearing neurons have been said to be devoid of normal microtubules ([114, 135] quoted by ). It has also been shown that the concentration of stable tubulin in tangle-bearing neuron was much decreased .
Types of accumulation
Numerous antibodies directed against various epitopes of the tau protein label the NFTs. The most commonly used are the two monoclonal antibodies Alz50 and AT8. Alz50 IR is conformational, requiring both two discontinuous epitopes, one at the N-terminus of the molecule, the other in the microtubule binding region . AT8 recognizes phosphorylated epitopes at serine 202 and threonine 205, or at serines 199 and 202 or at serines 205 and 208 . AT8 was shown to be particularly robust in inter-laboratory assessment [4, 42].
NFTs are principally found in the medium-sized pyramidal neurons of the hippocampus, of the entorhinal cortex and of layers III and V of the isocortex. The granule cells, e.g. of the dentate gyrus or of layer IV, are largely spared. So are the giant pyramidal cells of Betz, in area 4. The neurons of the fused pyramidal layer (III + V) that crosses the thickness of the cortex in the transentorhinal area and the pyramidal neurons of entorhinal layer II are thought to be among the first neurons to be involved . Tangles are also seen in the olfactory bulb at an early stage of the disease . Before being aggregated, abnormally phosphorylated tau fills the neuronal soma (so-called “pretangle” ).
On the other hand, extracellular NFTs, keeping the shape of a cell body, are frequently found in severely affected regions such as the entorhinal cortex and the pyramidal fields of the hippocampus; they are clear evidences of the neuronal death that is probably the common end point of neurofibrillary degeneration . These ghost tangles lack the carboxy-terminal sequence of tau . They may be associated with a cluster of neurites, that form a tangle associated neuritic cluster (TANC)  and simulates a neuritic plaque.
Many, but not all NFTs are ubiquitinated [157, 257] and labeled by p62 antibodies . Ubiquitination is probably a late event in NFT formation. NFTs are labeled by antibodies directed against prothrombin and thrombin , heparan sulfate proteoglycan , Fe65 , flotillin—a marker of rafts . They are also labeled by antibodies directed against various kinases such as CDK5 (see previously) or the active form of glycogen synthase kinase 3 β (GSK3β) .
Ultrastructurally, NFTs appear as paired helical filaments  or, as suggested more recently, as helical  or twisted ribbons . The filaments are made of tau in a cross β configuration .
The link between NFTs and amyloid deposits is still to be discovered: APP transgenic mice do not produce tangles unless a human, mutated tau gene is also transferred . A physical binding of unphosphorylated tau and Aβ peptide has been postulated as the initial step followed by tau phosphorylation and aggregation .
Tau antibodies as well as Gallyas silver stain show in AD brains a large number of small, fragmented, tortuous processes, weaving between the cell bodies. These “neuropil threads”  or “tortuous fibers”  contain PHF that accumulate in dendrites of tangle-bearing neurons . Neuropil threads invariably accompany NFTs and neuritic plaques. They occur at an early stage of the neurofibrillary degeneration . Their number is linked with the density of the NFTs and may decrease in the most advanced cases .
Neuritic corona of the senile plaque
As already mentioned and discussed, the neuritic corona of the senile plaque contains processes that are tau immunoreactive while another subset of the neurites is chromogranin A and APP positive. It has been shown in APP transgenic mice that dystrophic neurites in contact with the plaque remain continuous and connected to viable cell bodies .
Evolution of the neurofibrillary changes
Using cases at different stages of the disease and thick sections Braak et al.  and Sassin et al.  have reconstituted the natural history of the neurofibrillary changes within a neuron. They have confirmed the initial observation of Bancher et al.  that a diffuse labeling of the cell body and of the processes of the neuron was the first change to be detected (group 1 neurons). Tau protein secondarily aggregates in the processes which become tortuous (group 2 neurons). In group 3 neurons, the tau-positive dendrites appear fragmented and a NFT form in the cell body. Coarse AT8 immunoreactive granules are characteristic of the “early” ghost tangles (group 4) while typical, but AT8 negative, ghost tangles define group 5.
The combined use of AT8 and ubiquitin antibodies, as well as thiazin red (that labels fibrillary structure) and Gallyas silver technique on hippocampus pyramidal neurons has also suggested a maturation of tau-positive inclusions . They are first recognized by the AT8 antibody, then by the Gallyas method. Thiazin red recognizes only the fibrillary structures. Sixty percent of the NFT are ubiquitin positive. All the ubiquitin-positive NFTs are also stained by thiazin red : tau aggregation seems to be recognized by the ubiquitin–proteasome system only when it is fibrillary.
Topography and progression of tau accumulation
Contrarily to Lewy bodies which do not accumulate in the brain and thus probably disappear with the neurons that bear them , neurofibrillary pathology is not cleared—or only very partially—from the brain where it can stay in the extracellular space without causing an inflammatory reaction. The persistence of the NFTs has two main consequences: (1) The density of the alterations is correlated with the severity of the disease (2) It is possible to track the progression of the lesions by analyzing cases at various stages of the disease (see next paragraph). The way by which tau pathology progresses in the brain is not well understood. It is, however, clear that the progression is related to the anatomical tracts. An exceptional clinico-pathological case study has revealed that Aβ may accumulate—but tau pathology does not progress—when a disconnection has taken place: a small piece of cortex had been accidentally isolated from the rest of the brain following a surgical operation years before the development of AD. In this disconnected piece of cortex, Aβ had accumulated but there was no NFT, no neuropil threads and no neuritic plaque . This observation as well as the early involvement of neurites and the delayed formation of NFT (see the paragraph “Differential distribution of the deposits with and without neuritic corona” and ) favor the hypothesis of an extension of tau pathology through anatomical connections leading to a hierarchical distribution of the tau pathology tightly linked to clinical manifestations. This hierarchical distribution is the basis of Braak neurofibrillary stages.
Braak and Braak  have formalized the progression of neurofibrillary pathology in the cerebral cortex. The staging procedure is based on the topographical distribution of neurofibrillary tangles and neuropil threads. The trans-entorhinal and entorhinal cortex (stages I and II), then the hippocampus (stages III and IV) and finally the isocortex (stages V and VI) are sequentially involved. The study of a large cohort of cases  allowed computing the ages of maximal incidence and prevalence of the stages. The lesions occur earlier than generally thought (for stages I and II, age of maximal incidence: 47.5 years; maximal prevalence: 67.5 years) but progress over decades (for stages III and IV, age of maximal incidence: 87.5 years, of maximal prevalence: 97.5) .
The simple scheme of Braak stages has been generally adopted by the neuropathologists and has found no real competitor. It has been shown that the agreement between the observers was better when the lesions were abundant than when they were sparse [4, 100]. Braak staging of neurofibrillary pathology has been adapted to routine pathology  and is commonly used. It appears well correlated with the clinical status . Exceptions to Braak stages have been actively looked for; they are said to be frequent but appear generally minor . They were, however, thought to be sufficiently meaningful by Gertz et al.  to discuss the general validity of the hierarchical model. Braak stages involve the cerebral cortex but other regions are affected.
Olfactory bulbs, olfactory epithelium
Tau pathology has been well documented in the olfactory bulbs [23, 323]. Threads and NFT were found in cases at early Braak stages: tau pathology is present in roughly one-third of the cases at Braak stage II and in all the cases at stages V and VI . The initial report of tau positive changes in the olfactory mucosa was not confirmed by subsequent studies . Tau does not accumulate in the retina.
The subcortical nuclei are not considered in the basic scheme of Braak staging. They are however involved quite early in the progression of the disease and their lesions have important clinical consequences. It has been shown that the nucleus basalis of Meynert (innervating the isocortex in acetylcholine) and the locus coeruleus, at the origin of the noradrenergic fibers of the cortex, were both involved early, as soon as stage I [27, 140, 278]. Thickened cholinergic fibers and ballooned terminals are present even in non-demented middle-aged cases. Tangles and pretangles are found in the nucleus basalis even before involvement of the entorhinal cortex in control or MCI cases [121, 228]. The neuronal loss is severe in both the nucleus basalis and the locus coeruleus [18, 118].
In the hypothalamus, the pathology seems to have multiple meanings: neurofibrillary pathology is found in a small proportion of cases in the supra-optic and paraventricular nuclei, without correlation with the involvement of the medial temporal areas  and without neuronal loss . By contrast, neuronal loss is marked in the suprachiasmatic nucleus . The dysfunction of the network connecting the retina, the suprachiasmatic nucleus and the pineal gland could explain the low level of melatonin in AD patients . Neurofibrillary degeneration found in the neurites in contact with the vessels of the mediobasal nucleus of the hypothalamus in old males does not appear to be related to AD . In the nucleus tuberalis lateralis, many somatostatinergic neurons may be AT8 positive but with a variable amount of Aβ pathology in the neuropil. Most of the AT8 positive neurons are Gallyas negative .
In the thalamus, the intralaminar nuclei are affected at early stages. In these nuclei, the neurofibrillary changes not only involve the ascending reticular activating system and diverse motor or oculomotor systems, but also the medial pain system . Large neurons of the striatum are affected by neurofibrillary changes; neuropil threads may be numerous in the striatal gray matter [45, 254]. The nucleus accumbens, the olfactory tubercule and the tail of the caudate nucleus appear particularly vulnerable .
Neurofibrillary tangles may also be found in the substantia nigra, associated with a variable cell loss which is more marked when Lewy bodies are also present . The extrapyramidal symptoms, frequent in AD, have been correlated with the substantia nigra tau pathology . In a more recent study, they were found associated with the neuronal loss and only weakly with α-synuclein pathology; they were not linked to the accumulation of tau protein .
The nuclei of the pontine parabrachial region (medial and lateral parabrachial nuclei; subpeduncular nucleus) together with the intermediate zone of the medullary reticular formation are also affected early and the severity of the lesions increases with increasing stages of the disease . Neurofibrillary pathology occurs in the dorsal raphe nucleus at a very early stage; the other raphe nuclei are subsequently affected. These lesions explain the serotoninergic deficit found in AD. To our knowledge, the clinical correlates of these changes are not yet known.
Early cortical involvement
The cortical areas are probably not involved in the same hierarchical sequence in all the cases. If the neurofibrillary pathology is indeed the best correlate of the symptoms (as generally agreed, see below), then it should involve some specific areas in the unusual cases with focal onset (see “Focal deficit”). Almost one-third of the cases with primary progressive aphasia (PPA) were found to be related to AD . Munoz et al. have described the presence of “argyrophilic thorny astrocyte clusters” (ATAC) in seven out of eight cases with PPA associated with AD. ATAC could thus be a marker of this special type of AD . It may be predicted that a focus of neurofibrillary changes involves the language areas in AD-associated PPA. In the same way, the patients who suffer initially from disorder of higher visual function (so-called Benson syndrome) probably develop at an early stage neurofibrillary changes in the visual associative cortices. It is noteworthy, in this regard, that area 19, a visual association cortex, was found to be severely affected by neurofibrillary pathology in cases with an apparently normal cognition or with mild cognitive impairment. The involvement of area 19 and of the entorhinal/hippocampal cortices was not correlated. Lesions were occasionally found in the absence of substantial pathology in the hippocampus or entorhinal cortex .
Validity in the oldest old?
Several studies have shown that the correlation of the Braak (neurofibrillary) stages are less well correlated with the clinical data in the oldest old than in the younger population for two reasons: even low Braak stages may be associated with severe dementia; the frontier between clinically unimpaired and demented cases is blurred [120, 126].
Relationship between Aβ and tau pathology
Since mutations of APP and of presenilins cause familial AD (FAD) with abundant tau pathology while mutations of tau cause fronto-temporal lobar degeneration (FTLD) without Aβ peptide accumulation (for review see ), it has been suggested long ago that an alteration of APP metabolism inducing an increase in Aβ secretion was the trigger of a series of mechanisms leading ultimately to tau pathology and neuronal death. The recent finding that Aβ oligomers (or amyloid beta-derived diffusible ligands, ADDL) promote the phosphorylation of tau in primary cultures of hippocampus or in neuroblastoma cells  indeed suggests that tau pathology is secondary.
Several facts, however, indicate that this hypothesis may not apply to sporadic cases: in a large study analyzing the prevalence of the amyloid and neurofibrillary lesions as a function of age, Braak et al. showed that the latter preceded the former by several decades ; in other words tau pathology did not appear to be secondary but initial. In APP transgenic mice, abundant Aβ accumulation may be observed in the absence of tau pathology although mutations of the tau gene, when associated, induce neurofibrillary pathology that interacts with Aβ . Why tau and Aβ pathologies are so regularly and intimately associated, remains one of the major unresolved questions of AD physiopathology. The senile plaque, where Aβ deposits and tau positive processes meet, is probably one important site of interactions: the cytoplasmic domain of APP, phophorylated on threonin 668, could be the intermediate since it appears to be associated both with tau and Aβ , but there are also well-documented data indicating that Aβ and tau could meet at the synapse. Fein et al. prepared synaptosomes from cryopreserved material, labeled them with anti-Aβ and anti-phospho-tau fluorescent antibodies. The synaptosomes were analyzed by flow cytometry. Phospho-tau and Aβ were colocalized in nearly 25% of them .
This neuronal intracytoplasmic vesicle, a few microns in diameter, contains a central basophilic and argyrophilic granule. At electron microscopy, the vesicle is bound by a unit membrane. Granulo-vacuolar degeneration (GVD) is seen most frequently (although not exclusively) in the pyramidal neurons of the hippocampus, usually in association with NFT. The granule of the GVD is immunolabeled by antibodies directed against tubulin , ubiquitin  and neurofilaments . Anti-tau antibodies label GVD regularly [89, 91, 143].
Granules, 1–5 μm in diameter, apparently attached to the soma of the pyramidal neurons of the CA1 sector of the hippocampus may be seen in Alzheimer and Pick disease cases. They are positive for the glutamate receptors GluR1 and R2 and are also labeled by anti-ubiquitin antibodies [20, 265]. They are more frequently seen in pre-tangle neurons than in neurons containing a fully developed NFT or in normal neurons. By immunoelectron microscopy perisomatic granules (PSG) were found to consist of GluR1-2-reactive enlarged synaptic boutons containing tubulo-filamentous or floccular material .
These rod-like, brightly eosinophilic intracytoplasmic structures are principally observed in neuronal processes of the CA1 sector of the hippocampus. They appear as paracrystalline structures at electron microscopy and are labeled by a variety of antibodies directed against actin, actin-associated proteins, tau, neurofilament, and the C-terminal fragment of APP. They are frequent in AD but far from specific; they have been initially described in cases with amyotrophic lateral sclerosis and parkinsonism–dementia complex in Guam, but are also found in controls and Pick’s disease (for review, see ). They are more frequent in AD than in controls .
The assessment of the density (or better of the total number) of neurons in the cortex or in the subcortical nuclei is fraught with bias if the counting of the profiles of the cell bodies is performed without correction. The large cells have indeed a higher probability than the small ones of being incorporated in the microscopic section. The large neurons are therefore overrepresented. This overrepresentation is proportional to the size of the object that is counted: atrophy of the cell body would therefore decrease the number of profiles per unit area and lead to the wrong conclusion that neurons have been lost when they are only smaller (“pseudo-loss” ). Small nucleoli (around 2 μm in diameter), less frequently re-cut, are less subjected to bias than large cell bodies. There would be no bias, if each neuron to be counted (in 3 dimensions) was transformed into a point that could not be sectioned (i.e. of dimension 0). There would be no “re-cut” with the consequence that the number of points counted in a slice could be directly extrapolated to the number of neurons per unit volume. The di-sector technique  consists in taking the extremity of the cell as the point that is counted: the cell is taken into account only if it is visible on the first, look-up, section and absent from the second one, i.e. if the extremity of the cell is located between the two sections. Other methods have been devised to evaluate the local variation in cell density: in Dirichlet tessellation method, it is the free surface around each cell that is measured (which is inversely proportional to the density). By this method, it is possible to evaluate the local density and to measure its variation . Using these different methods, it has been found that neuronal loss in AD was focal. The global number of cortical neurons is not decreased to a sufficient extent as to override the very high interindividual variation . Focally, severe neuronal loss has been documented in layer II of the entorhinal cortex where the loss may reach 90% of the normal neuronal population in the most advanced cases , in the CA1 sector (where the number of neurons was reduced threefold in Alzheimer cases) , in the superior temporal gyrus  and in the supramarginal gyrus . The neuronal loss may be a late phenomenon with respect to the neurofibrillary pathology, but in the regions where the tangles occur at a presymptomatic stage, the neuronal loss may be found quite early in the course of the symptomatic disease: Kordower et al. compared the number of neurons in layer II of the entorhinal cortex in a population of prospectively assessed aged controls, patients with MCI and early AD. Atrophy of layer II, related to neuronal loss, was significant in patients with MCI (a loss of 64% of the total number of neurons) but the difference was not significant between MCI and AD cases as if neuronal death waned with time . The hippocampus pyramidal neurons expressing secretagogin, a calcium-binding protein, were found to be resistant, a resistance that may be related to the scarcity of tau accumulation in these neurons . Dirichlet tessellation has shown that the neuronal loss predominated in layers II and III in the parietal cortex and could reach 8 millions in the parietal lobe of AD cases . In the cortex, the large and middle-sized pyramidal neurons are the most vulnerable. The very large Betz cells and the small granule cells are relatively spared.
the olfactory bulb and the anterior olfactory nucleus ,
the medial part of the substantia nigra (SN) (controversial see  neuronal loss predominating in the lower and medial part of the SN and : no neuronal loss). A severe neuronal loss in the SN should anyway suggest another diagnosis than AD,
the rostral part of the locus coeruleus (from which the cortical projection arises) ,
the serotoninergic raphe nuclei .
The cause of the neuronal loss remains a subject of controversy: ghost tangles indicate that neuronal death is at least associated with, but possibly linked to, the presence of NFT. The calculation made by Cras et al.  lead to the conclusion that the number of neurons lost is close to the number of ghost tangles, implying a direct relationship. Gomez-Isla et al. as well as Kril et al. reached the opposite conclusion: the neuronal loss was said to be in excess of neurofibrillary tangles “by manyfold” [129, 188]. Grignon et al. found that the threshold which best discriminated between the cases with and those without significant neuronal loss was a density of NFTs higher than 5/mm2 .
It has become commonplace to state that apoptosis, related to Aβ toxicity, was the cause of neuronal death in AD. If there are indeed data that show that apoptosis may take place in cell cultures under the effect of Aβ peptide, the evidences in human neuropathology need to be critically assessed. The technique of in situ end labeling (ISEL) labels a large number of neuronal and glial nuclei in AD cases [95, 195] but their number is too high to be taken as direct evidence of cell death: if, to take a quantitative example, 5% of the neurons in the cortex died per year, then the number of surviving neurons after 1 year would be 95%, 95% × 95% after 2 years, and, generally (95%)n after n years. With this value the number of surviving neurons after 10 years of progression would be 0.9510 = 60%, a value that would be considered very high in view of the assessments of the neuronal loss reviewed in the preceding paragraphs. What would then be the frequency of the cells with apoptotic morphology in a microscopic slide if as many as 5% of neurones would die per year? Their number would obviously depend on the delay between the first morphological sign of apoptosis and the disappearance of the cell. If the dying process took 1 day, then 5%/365 days of the neurons would be seen with signs of apoptosis i.e. 0.0001 = 1 cell over 10,000. Since the number of cells positive for ISEL is much higher, two solutions must be considered: either ISEL-positive cells are not undergoing apoptosis or the time during which the signs of apoptosis may be detected is much longer than 1 day. The first conclusion was drawn by Stadelmann et al.  who suggest that DNA is fragile in AD and degraded postmortem, rather than actually broken in vivo. As Lucassen et al. , they emphasize the scarcity of apoptotic nuclei and the absence of oligonucleosomal laddering aspect on Southern blots. Moreover, the density of ISEL positive cells is not linked to the density of the lesions surrounding them . One in 1,100–5,000 neurons were immuno-positive for caspase 3, an enzyme activated during apoptosis, a frequency more in agreement with the assessed neuronal loss . Caspase 3 may cleave APP770 in its cytoplasmic domain between histidine 732 and 733, creating a neo-antigen. An antibody recognizing this neo-antigen labels a large number of neurons in samples from hippocampus and frontal cortex of normal cases. The labeling was much stronger and more diffuse in AD cases. The connection between this caspase-cleaved APP and apoptosis is not known .
Anderson et al. used ISEL technique in Down syndrome (DS), AD and control cases. Their results are at contrast with the previous ones: they found a high density of ISEL-positive cells in controls; their density was higher in AD and DS cases. Apoptotic nuclei were seen and pulsed-field gel electrophoresis found laddering in 50% of the cases with AD and in 71% of the cases with DS . These results were not confirmed by Raina et al.  who found that the initial stages of apoptosis were not followed by the expected cascade of proteolysis; these authors believe that AD is an exceptional example of “abortive apoptosis”. Broe et al.  using more restrictive criteria to accept a nucleus as “abnormal” also considered that the frequency of apoptosis was high specifically in AD where it reached 23%. C-Jun, a protein associated with the activation of the apoptotic cascade, was expressed in AD  and found associated with NFTs . In summary, there are evidences that point to apoptosis in AD. However, there is no good way yet to determine neither the prevalence of apoptotic cells in AD brain nor the kinetics of the neuronal death. The neuron, being post-mitotic, could well survive despite DNA-damage. As mentioned earlier, re-entry is one of the possible triggers of neuronal death.
Synaptic pathology in AD has two aspects: (1) synapses participate to senile plaques  (2) the total number of synapses in AD brain decreases with time. This topic is principally dealt with in this paragraph. The loss of synapses has been described ultrastructurally and with immunohistochemistry. Scheff et al. [78, 280, 282] showed that the number of synapses was decreased but that, in several instances, the apposition length was maintained. This finding could indicate that the synapses are less abundant but larger in AD. The count of the synapses at the ultrastructural level is decreased at an early stage in the CA1 sector . Pre-synaptic markers (such as synaptophysin and SNAP-25) or post-synaptic ones (such as PSD-95) have also been used to analyse the synaptic alterations. Synaptophysin was found to be decreased at early stages of the disease , leading to the conclusion that synaptic loss was the best correlate of the intellectual deficit . The reality seems more complex: tau pathology appears better correlated with the cognitive deficit than the drop in synaptophysin IR [87, 238]. Moreover, other presynaptic markers such as SNAP25 do not seem to be decreased [86, 238]. Generally speaking, the vesicular markers (synaptophysin for instance) are decreased while membrane markers (SNAP-25) are maintained . Finally, PSD-95, a post-synaptic protein was found to be paradoxically increased in the frontal cortex .
The morphological basis of this decrease in synaptic markers is complex: the decrease is, for instance, not accentuated in diffuse deposits ; clusters of synapses have been seen along neuropil threads . However, these synapses may not be functional. Major changes take place in contact with the amyloid deposits. Abnormalities of dendrites, the most frequent being loss of dendritic spines, are found in close association with deposits of fibrillar Aβ in man as in APP transgenic mice . In man, there is an inverse correlation between the dendritic arborisation  or the dendritic arborisation index  and the density of NFTs in the subiculum. One study resorted to synaptosomal preparation and flow cytometry. The most striking finding was a very high increase in the Aβ content of the synaptosomes from AD patients . In vitro experiments indicate that Aβ oligomers bind to post-synaptic densities of presumed excitatory pyramidal neurons  and rapidly inhibit LTP . Moreover, Aβ 42 was found to accumulate in multivesicular bodies in pre- and post-synaptic compartment . It should be added that it has recently been possible to follow in vivo the changes in the morphology of dendritic spines in transgenic APP mice .
In conclusion, even if the decrease in synaptophysin may not be the best correlate of the intellectual deficit, there are numerous evidences that the synapses play a major physiopathological role in AD although the timing of the lesions is not yet fully understood: they contain Aβ peptide and Aβ oligomers bind to them. The number of synapses, evaluated by multiple techniques, has been found to be decreased and it could be the major final common pathway of the pathology.
The neuropil of AD cases may sometimes appear vacuolar and, at times, the status spongiosus may raise diagnostic difficulties with Creutzfeldt–Jakob disease . Such vacuolization is more frequent when AD is associated with Lewy pathology . It is to be distinguished from the band of spongiosis that is the consequence of de-afferentation: the band of spongiosis seen in the molecular layer of the dentate gyrus, for instance, is due to de-afferentation  from the entorhinal neurons of the pre alpha layer severely affected by neurofibrillary pathology .
Macroscopy in Alzheimer disease
The microscopic changes that were reviewed in the preceding paragraphs help to understand the macroscopic findings. The atrophy that is characteristic of AD involves the areas where the tau pathology predominates  as well as the neuronal loss: the entorhinal cortex, the hippocampus and the amygdala. The inferior temporal and the superior and middle frontal gyri are severely affected while the inferior frontal and the orbitofrontal gyri are spared . It is not known if synaptic loss or loss of processes may cause atrophy without neuronal loss but a strong correlation has been established between neuronal loss and atrophy in AD . Atrophy is not correlated with the amyloid burden .
The loss of volume is both related to a decrease in the thickness of the cortical ribbon and to its length. Measurement of the thickness can now be performed automatically on MRI views . The decrease in the length of the cortex is due to loss of columns perpendicular to the surface, neurons but probably also fibers —a disruption of columnar arrangement that has also been described microscopically . Focal atrophy has been observed in AD in the cases in which amnesia was not the initial symptoms (see below “Focal deficit”).
The white matter contains the axons of the neurons that degenerate in AD. This is why atrophy of the white matter is correlated with the gray matter atrophy [187, 214] and does not seem to exceed that observed in normal aging , although it has sometimes been considered that it could precede and be of a larger magnitude than gray matter atrophy . Severe amyloid angiopathy may be associated with a severe leukoencephalopathy that is probably the consequence of the synergistic effect of degeneration and ischemia . Certain long fiber tracts atrophied specifically: the corpus callosum is smaller in its temporo-parietal segment . More generally, modern imaging techniques have been able to show that interhemispheric fiber tracts were selectively affected whereas extracortical projecting fibers were spared .
Inflammation: relation with immunotherapy
Activated microglia , early components of the complement cascade , proinflammatory cytokines  are evidences of an ongoing inflammatory process in the senile plaque , whose role and importance have been discussed. The P component is a protein found in the serum and which is associated with many types of amyloidosis. The C1q complement factor and the P component are associated with fibrillar Aβ and could play an important role in the clustering of microglial cells and the secretion of cytokines in the senile plaque . Diverse views have been expressed concerning the significance of AD inflammation: it has been considered toxic or, on the contrary, useful . Active or passive immunization against various epitopes of the amyloid peptide has been shown to reduce the amyloid load  (but also to increase atrophy and ventricular volume  with little clinical improvement ). The removal of Aβ is associated with phagocytosis of Aβ by macrophages . On the other hand, the activation of microglia, seen in multiple sclerosis, does not prevent the formation of AD lesions .
The packing density of fibrous astrocytes is increased in the cortex of AD patients . GFAP concentration, measured by ELISA, is massively increased (more than 10 times the normal values) in the cortex, thalamus, brainstem and even cerebellum (in which pathology is limited to Aβ deposition in the most severe cases) . The astrogliosis is correlated with the density of NFT and of Aβ deposits . As mentioned earlier, the senile plaque is wrapped in astrocytic processes . Astrocytes in and around diffuse deposits may contain granules that are intensely labeled by anti-Aβ antibodies , probably the result of endocytosis. According to Akiyama et al. , most of these granules are negative for antibodies to the N-terminally located sequences of Aβ including 6E10 (Aβ AA 1–17) and 6F/3D (Aβ AA 8–17) while Thal et al.  detect Aβ with 6F/3D and 4G8 (Aβ AA 17–24) in subpial astrocytes. The astrocyte shown in Fig. 2c was also labeled with the 6F/3D antibody. These discrepant results may be related to the heterogeneity of the isoforms of Aβ found in the astrocytes. This heterogeneity, discussed by Akiyama et al.  is probably due to differences in the stages of degradation rather than to the type of Aβ that has been endocytosed (parenchymal Aβ was not N-truncated in the areas studied).
Plasticity and neurogenesis
It is accepted that AD pathology is not only characterized by the loss of normal structures but also by regenerative processes . Cajal  himself thought that the senile plaque was the consequence of an altered regenerative process. Aberrant dendritic sprouting has been noticed, with Golgi method, in the basal nucleus of Meynert where abnormal filopodia were seen  or in senile plaques where the axon collaterals appeared to branch richly as they entered plaques . “Pleomorphic outpouchings of terminal and preterminal” dendrites were also seen . The excess of GAP-43-positive fibers in areas of synaptic degeneration reflects the importance of regeneration .
More recently, signs of neurogenesis, taking place normally in the paraventricular zone and in the dentate gyrus of adults, have been studied in AD with controversial results: markers of neurogenesis have been found to be increased ; however, it is suspected that newly generated neurons do not mature into functional neurons .
Correlation with clinical data
The case–control type of study is difficult to apply to AD, since the control group, if matched for age, is invariably contaminated by pre-symptomatic cases . Since the seminal paper of Blessed et al.  numerous groups have thus attempted to correlate the quantitative assessment of the cognitive deficit with the density of positive lesions or the decrease in the density of normal constituents. Such procedure needs a prospective clinical assessment and implies that the evaluation of both the cognitive deficit and of the histopathology is reliable. It has been repeatedly shown that this evaluation is unreliable when comparing different laboratories and must also be critically taken for the cases with but a few lesions [5, 100, 233]. Moreover, the relationship between the current intellectual scales and the pathology is not linear and is difficult to model .
This said, tau pathology has been almost invariably found to be the most reliable correlate of the cognitive deficit in the numerous correlative studies that have been published [31, 35, 37, 98, 122, 128, 243]. It should however be stressed that the extension of neurofibrillary pathology (i.e. the number of involved areas) rather than its density may be the significant variable . It is noteworthy, in this regard, that a very high correlation between the clinical status and the Braak neurofibrillary stages has been mentioned several times , most noticeably in the Nun study . In most of the aforementioned studies, Aβ accumulation has been found to be correlated with the cognitive deficit, but less significantly than tau positive alterations; most of the studies mention cases in which a high density of diffuse deposits is not associated with low clinical scores. The central role of Aβ in Alzheimer pathology, therefore, does not seem to be confirmed by clinico-pathological studies. However, two remarks must be made: one concerns the methodology of the studies, the other, the role of a morphologically invisible pool of Aβ, a pool that can only be detected by Western blot or ELISA. The quantification of plaques must be based on a careful sampling to be reliable: there is indeed a large local and regional heterogeneity. Bussière et al.  suggest that some unexpected results could be explained by an inappropriate sampling scheme of the regions in which the counting was performed. They found a correlation between clinical status and total Aβ accumulation in the entorhinal cortex and in the subiculum but there was no correlation in the superior frontal gyrus. Since Aβ and tau deposits are correlated, one may hypothesize that the link with the cognitive deficit is more direct for one lesion than for the other. Since tau pathology appears better correlated with the clinical data, it is probably more directly linked with the symptoms than Aβ accumulation. However, the visible accumulation of Aβ peptide may not be the crucial variable: fibrillar Aβ, being highly insoluble, interact probably little with its environment acting as a “black hole” emitting no molecule able to reach neighboring cells. On the contrary, the soluble pool of Aβ made of diffusible molecules may reach membranes and specific receptors to play a toxic role. It has been shown that increase in Aβ concentration preceded tau accumulation and was higher in advanced cases than in MCI or cognitive normal patients . McLean et al.  have found a good correlation between the soluble fraction and the clinical status, a correlation that was not found with the insoluble pool. Further studies have refined the analysis of the soluble pool. By using a detergent to permeabilize the cell membrane, Steinerman et al.  provide evidences that the fraction of the soluble pool that is mainly toxic is intracellular.
It is highly plausible that the neuronal loss and the intellectual deficit are also correlated but the interindividual variation in the number of neurons is such  that it has, to our knowledge, not been definitely ascertained, neuronal loss having been usually evaluated in case–control types of study. We have already discussed the correlation between the synaptic loss and the intellectual deficit (see above).
The clinico-pathological correlation is less significant in the oldest-old population. Haroutunian et al. stratified their population of 317 cases in young-old, middle-old, and oldest-old subgroups. While the density of NFT and neuritic plaques increased more than tenfold as a function of the severity of the disease, such correlations could not be found in the oldest-old subgroup, mainly because the lesion density was lower at that age .
AD type lesions have been often found, sometimes in large number, in patients without clinical symptoms. This observation explains why it is sometimes stated that a moderate number of plaques and tangles is “normal” in aged patients. The opposite opinion—that all lesions of AD type are pathological—has been strongly stated by the experts who designed the NIA Reagan criteria : in this view, AD has an initial asymptomatic phase of long duration and a symptomatic phase, during which the clinical diagnosis of AD or of MCI may be made. The threshold between the normal and symptomatic phases depends on many variables. Knopman et al.  studied 39 longitudinally followed, cognitively normal elderly individuals. They concluded that “the majority of individuals who are cognitively normal near the time of their death have minimal amounts of tau-positive neuritic pathology (Braak stage < IV and neuritic plaques < 6 per 100× field in the most affected neocortical region)”. Markesbery et al.  found that the density of neurofibrillary tangles in the medial temporal lobe (entorhinal cortex, subiculum, CA1 and amygdala) was the most efficient variable to draw the border between the normal and the MCI (amnestic type) populations. The density of neuritic plaques and of neurofibrillary tangles was on average higher in early AD cases than in MCI patients and “from a neuropathologic perspective, it appears that amnestic MCI is, in reality, early Alzheimer disease”. In a similar way, Bennett et al. found, in their “normal” population, cases that met the criteria for AD but, with refined testing, concluded that, even in the so-called pre-symptomatic phase, AD lesions (in the intermediate or high probability range of the NIA Reagan criteria, see later) were associated with subtle alterations of memory . It is interesting to note that in both studies by Knopman et al. and Markesbery et al., the diffuse deposits (“diffuse plaques”) are not a discriminative variable. As we shall see later, the lesions’ threshold that causes symptoms is lowered by associated alterations (especially vascular) and in the oldest old .
The first formal neuropathological criteria that were internationally recognized were those published under the name of Khachaturian . They were strongly biased toward the concept that the senile plaques rather than the neurofibrillary tangles were the characteristic lesions of AD. This concept was fueled by the finding that many unrelated pathologies were characterized by the presence of neurofibrillary tangles that were, therefore, thought unspecific . Khachaturian criteria proposed a sampling scheme and suggested cut-off quantitative values of senile plaques density, modulated by the age of the patient and leading to a binary, “black and white”, diagnosis of AD. Retrospectively, these criteria may appear insufficient on several aspects: the threshold of the senile plaque number had been conventionally determined and was not based on any epidemiological evidences or theoretical concept. Moreover, the “threshold” approach implied a precision in the measurement that could not be reached then as well as now. Several studies suggested that inter-laboratory reproducibility in the neuropathological protocols was not sufficient to base diagnostic criteria on quantitative values [100, 233]. By contrast, the agreement was much better when the ranking of the cases according to the severity of the lesions was left to the observer. As more recently confirmed, the assessment was much less reliable when the density of the lesions was low than when it was high [4, 100]. It is, in other words, more difficult to ascertain that there is no AD pathology than to assess full-blown lesions. The subsequently developed CERAD protocol  recommended a semi-quantitative assessment of the senile plaques and added the categories “CERAD neuropathologically possible” and “probable” AD between the binary result of Khachaturian criteria, “normal brain” and “definite Alzheimer disease”. These new categories introduced a probabilistic approach to the neuropathological diagnosis that was quite new at the time.
The finding that plaque counts could be high in subjects with minimal cognitive impairment and that tau pathology was the best correlate of dementia led to the NIA Reagan criteria  which formalize the relationship between clinically ascertained dementia and the density of amyloid as well as tau pathology. Dementia was related to the lesions with a “high”,” intermediate” or “low likelihood”. The possibility that lesions could have been observed in the absence of clinical signs was left for further studies. Recent developments have shown that the focus on diagnosis might have been misleading for several reasons: since the lesions of Alzheimer disease do not necessarily involve contiguous areas, apparently normal nervous tissue borders areas with severe involvement. As a consequence, a scientist can receive a minimally, or even at times, “normal” sample from a so-called AD brain. On the other hand, Alzheimer pathology is much more prevalent than once thought; it is rare, for instance, to observe lesions-free entorhinal cortex or hippocampus in aged subjects. Finally, recent studies have shown that Alzheimer pathology constitutes but one variable in the often complex equation that leads to dementia  with the consequence that the evaluation of a demented case should take into account the various aspects of pathology. Consequently, rather than to propose a binary diagnosis, the neuropathologist must determine the stage—i.e. the topography of the neurofibrillary lesions  and the phase—i.e. the topography of the amyloid deposits . When one sample is assessed for research purpose, it may also be useful to evaluate the grade—i.e. the severity of the lesions in the given sample . A similar assessment of the Lewy type pathology and of the vascular changes should be performed to catch the complexity of the “real” pathology that is rarely pure. The recent McKeith criteria attempt such a combined assessment. The experts recommend evaluating separately Alzheimer and Lewy pathologies and to report them in a 3 × 3 table. The high, intermediate and low probabilities of AD (Reagan-National Institute on Aging criteria) constitute the columns of the tables, while the Lewy pathology (brainstem predominant, limbic and diffuse neocortical) constitutes the lines. A high, intermediate or low probability of explaining the symptoms by the Lewy pathology is attributed to the nine intersections of this 3 × 3 table .
Various types of AD
Although the monotony of AD neuropathology has generally been emphasized, some pathological and clinical variations have been described. Their number should increase with the scrutiny of the observation and the size of the cohorts.
These cases were defined as “lacking neocortical neurofibrillary tangles or having very few” of them. A significant proportion of these cases were found to also have brainstem and cortical Lewy bodies . The phenotype of plaque-only case does not seem to be very different from the common type : they were older, less impaired, and progressed more slowly. It is worth mentioning here that the second case of Alois Alzheimer was said to be of the “plaque-only” type .
As already mentioned the presence of amyloid deposits in the neocortex and of neurofibrillary tangles confined to the hippocampus (Braak III–IV; Thal phase 1) constitutes one of the phenotype in the formalized progression of AD. Those cases, when occurring without associated pathology, usually survive and reach more advanced stages. The plaque-only phenotype is related, in this view, to the poor prognosis of even these early lesions when associated with Lewy pathology.
These cases are old (80–90). Dementia is associated with numerous neurofibrillary tangles in the allocortical areas (entorhinal region, subiculum, CA 1 sector of hippocampus, amygdala) with no or only few diffuse plaques in the isocortex, no isocortical tangles and no neuritic plaques [32, 170]. 3- and 4R tau are present in the tangles as in the common form of AD . The prevalence of the ApoE4 allele is low in this subset of patients in contrast to that observed in the common form of AD .
Early onset AD and neuropathology of cases with identified mutation
The severity of the lesions (including the density of neurofibrillary alterations, the amyloid load and the cortical atrophy) is much more marked in early onset AD than in the common form with onset after 70 years of age. The atrophy of the medial temporal lobe is more severe. In presenilin 1 mutations, the fronto-temporal atrophy is increased .
Genetic cases are characterized by an early onset and severe lesions. Some of them may suggest definite mutations (for references see: Alzforum (http://www.alzforum.org) or the AD and FTD mutation data base curated by Marc Cruts (http://www.molgen.ua.ac.be/ADMutations): as already mentioned, cotton wool plaques, although not being specific , are commonly seen in some mutations of the presenilin 1 gene (PSEN1) . They may be associated with spastic paraplegia that may precede dementia for years. No definite mechanism explaining those phenotypic particularities has been elucidated . Ataxia has also been one of the presenting symptoms in a few families with the PSEN1 Pro117Ala mutation .
The predominance of the amyloid angiopathy is a known characteristic of various mutations of the APP gene such as the Dutch mutation (Glu693Gln)  and the Iowa mutation (Asp694Asn)  or in the PS1 gene. Frequent occurrence of epilepsy, severe cerebral amyloid angiopathy causing hemorrhages in a significant number of cases, and intracellular accumulation of Aβ x-40 in the neurons of the dentate gyrus and of the pyramidal cell layer of the hippocampus are characteristic of the duplication of the APP gene .
The Apo ε4 variant, especially when present on the two alleles of the gene, is associated with early onset AD in which capillary amyloid (type 1) angiopathy is frequent .
While the impairment of the episodic memory is almost invariably the mode of onset of AD, a focal deficit may be the initial symptom as already mentioned (see “Early cortical involvement”): in primary progressive aphasia (PPA) due to AD, aphasia is generally of the “logopenic” type. The NFTs may be more abundant on the left side at autopsy (i.e. at a late stage of the disease). “However, the asymmetry is low and inconsistent” . All the seven posterior cortical atrophy cases (visual agnosia, Balint’s syndrome and alexia) in the series of Alladi et al.  were related to AD. AD was also the major etiology in the series of 21 neuropathological cases studied by Renner et al. . This clinical syndrome is probably to correlate with the finding of a focus of NFTs in area 19 in preclinical AD . In their series of AD with progressive focal syndrome Alladi et al., studied 12 cases of cortico-basal syndrome (progressive apraxia, alien hand syndrome, unilateral rigidity), among which 6 were related to AD .
With associated lesions
The association of Alzheimer pathology and of Lewy type pathology is common, probably more frequent than chance would predict. Initially, the Alzheimer pathology found in Parkinson disease was thought to explain the cognitive symptoms met in this disease . Nowadays, it is considered that the Alzheimer pathology is generally insufficient to warrant the diagnosis of AD in Lewy body diseases (i.e. in dementia of Parkinson disease and in dementia with Lewy body)  although around one-third of the PD cases with dementia had severe neuritic pathology in the series of Jellinger . The presence of Lewy bodies in cases of otherwise typical AD has been recognized. Lewy pathology has also been recognized in cases with APP or PSEN1 mutations [194, 201] or with trisomy 21 . The amygdala is a particularly vulnerable region and this has led to isolate AD with amygdala Lewy bodies , otherwise unremarkable.
Recent studies have shown that pure AD was not as frequent as previously thought especially in the community and in the old population. The MRC-CFAS study (an unselected, community-based neuropathology study in an elderly (70–103 years) UK population, ) as well as the Nun study (involving both normal and cognitively impaired, unselected persons ), have shown that vascular lesions were frequently associated and that this association could modify the clinical aspect of the disease. In the Viennese series, as many as 20% of the AD cases had also vascular pathology . Vascular and Alzheimer lesions have a synergistic effect [127, 352]. The vascular lesions lower the threshold at which AD lesions become symptomatic. The topography of the lesions has obviously a crucial importance on their cognitive impacts [127, 351].
The studies performed in the community have also shown that the correlation that was found between the intellectual status and the density of the lesions was not as strong as in institutionalized patients. In the MRC-CFAS study, vascular lesions and Alzheimer pathology were found in respectively 78 and 70% of the 209 autopsy cases . Dementia was present in 48%; 64% of those with a diagnosis of AD but 33% of the non-demented subjects had equivalent densities of neocortical plaques. The “reserve hypothesis”  has been formulated to explain major interindividual variation in the relationship between the severity of the lesions and their clinical counterpart: some individuals would have more neurons or connections able to sustain more lesions than others.
The loss of neurons, accompanied by intense astrogliosis, in the CA1 sector and in the subiculum may be found isolated in old patients but may also be particularly prominent in AD, which is the most common co-occurrence of hippocampal sclerosis . In a Viennese series, hippocampal sclerosis was found in 3.1% of the 650 autopsy-proven AD cases. The cases with hippocampal sclerosis were, as a mean, older and had a higher prevalence of coronary heart disease, suggesting that hypoxic–ischemic episodes may have been a causative factor . TDP-43 pathology could also favor the occurrence of hippocampal sclerosis (see below).
Argyrophilic grain disease
The lesions of argyrophilic grain disease (AGD) may be difficult to distinguish in advanced AD cases. Fujino et al. used a specific 4 repeats (4R) tau antibody to distinguish neurofibrillary lesions (4R and 3R tau positive) from grains (only 4R tau positive). The frequency of AGD in AD reached 26%. AD patients with AGD were significantly older; H1 tau haplotype was overrepresented and Apoε4 allele was not . Current literature does not permit to determine if AGD is indeed more frequent in the AD population or if the association AD-AGD is fortuitous, both diseases being frequent.
TDP-43 and AD
TDP-43, the protein that is phosphorylated, ubiquitinated and abnormally accumulated in the cytoplasm of neurons in some types of fronto-temporal dementia, has also been found in the dentate gyrus in as many as 34% of AD cases. These were clinically more advanced than the cases without TDP-43 inclusions. Medial temporal atrophy was more marked and the occurrence of hippocampal sclerosis more frequent [9, 175]. Caspase-cleaved TDP-43 was detected in many AD lesions (Hirano bodies, tangles, reactive astrocytes and neuritic plaques) . In cases with AD and Lewy pathology, TDP-43 could be partly colocalized with α-synuclein pathology but not with tau accumulation .
Our understanding of AD pathology has made important progress during the last 20 years: the components of the lesions have been identified; the clinical correlates of the changes in cohorts of patients or in the community have been thoroughly studied; the progression of the pathology has been described. New imaging techniques, new and powerful animal models have helped understanding the time course and the mechanisms of the lesions. However, some basic questions remain unsolved: the relationship between Aβ accumulation and tau pathology is still badly understood; the mechanism of sporadic AD is still debated. Increased production of Aβ peptide, defect in its clearance or initial disturbance of tau metabolism are valid hypotheses that await new neuropathological observation, postmortem and in vivo.
The help of the medical doctors, students and scientists who worked in the laboratory (Yi He, Toshiki Uchihara, Marie-Anne Colle, Pascale Lacor, Malika Bennecib, Yolanda Arends, Wienneke Metsaars, Nadège Girardot, Thibaut Lebouvier, Claire Perruchini among others) is greatly acknowledged: as well the expertise of the technical staff. Several studies mentioned in this work were supported by the grants from ANR (ChoAD) and LECMA.
Conflict of interest statement
The authors declare that they have no conflict of interest.