Accumulation of amyloid β-protein (Aβ) in the cerebral vasculature is a common pathological feature of Alzheimer’s disease (AD) and cerebral amyloid angiopathy (CAA) [1,2,3]. Many biochemically distinct types of amyloid have been reported to deposit in cerebral vessels such as Aβ, cystatin-C (Icelandic CAA), transthyretin, gelsolin, prion protein, and the ABri and ADan subunits in familial British and familial Danish dementia [3]. However, CAA caused by Aβ is the most common form of sporadic CAA and also occurs in AD patients [1,2,3].

Aβ is a 40 to 42 aminoacid peptide derived from the amyloid precursor protein (APP), a type 1 transmembrane glycoprotein, through sequential proteolysis by β-secretase and γ-secretase/presenilin complex. A prior cleavage by α-secretase prevents the generation of full-length Aβ and generates p3 (Figure 1). Aβ has a pronounced propensity to form fibrillar aggregates, and the longer, more hydrophobic Aβ42 is also more prone to form fibrils than the shorter Aβ40 isoform. According to the amyloid cascade hypothesis, Aβ deposition is causative for the pathogenesis of AD [4]. One of the most convincing evidence to support this premise is identification of mutations in APP near the β- or γ-secretase cleavage site or in the presenilin gene causing familial forms of AD that increase the absolute or relative levels of Aβ42 [4]. While in these familial forms Aβ is also deposited in vessels, the most severe CAA is observed in patients with APP mutations that lie within the Aβ domain such as Flemish, Dutch, Italian, Arctic, and Iowa, amongst others (Figure 1).

Aβ CAA (henceforth CAA) is shown to affect leptomeningeal and cortical arteries, arterioles, capillaries, venules and veins in both sporadic and familial CAA, and in almost all AD patients [5,6,1]. Depending on the predominant type of vessel involved, two types of CAA have been distinguished: CAA-type 1 and type 2. CAA-type 1 affects meningeal and cortical arterioles, capillaries, veins and venules while CAA-type 2, also called large-vessel CAA, affects meningeal and cortical vessels but has a tendency to spare cortical capillaries [7]. In CAA-type I, Aβ frequently permeates the glioadventitial junction of capillaries and small arterioles and extends into the parenchyma as “dyshoric angiopathy” [5,8]. Also, depending on the level of vascular involvement, a simple 3-tiered grading system of mild, moderate, and severe CAA has been proposed [9]. In “mild” CAA, focal Aβ deposition restricted to the tunica media occurs in an otherwise normal vessel. In “moderate” CAA, Aβ deposition involves the entire thickness of tunica media, thus completely replacing the smooth muscle cell layer. No evidence of recent or old blood leakage is found. In “severe” CAA, Aβ deposition involves other vascular wall layers and the vascular architecture is severely disrupted with secondary changes such as microaneurysmal dilatation, fibrinoid necrosis, and vessel rupture [9,10,11].

CAA also has important and sometimes fatal pathological consequences. One of the most well-recognized complications of CAA is spontaneous, often recurrent, intracerebral haemorrhage, usually involving the cortex and/or subcortical white matter [12,13,14,9,15,16]. Secondly, CAA also increases the risk of cerebral infarction, especially in the watershed areas, and is frequently associated with ischaemic leucoencephalopathy [17,18,19]. Thirdly, CAA has also been associated with progressive dementia in sporadic CAA patients [20,21,22]. Even in patients devoid of any AD pathology, CAA has been significantly associated with the presence of clinical dementia [23,22]. Interestingly, CAA-type I has been shown to be more closely related to AD pathology than large-vessel CAA [1,24], and increased perivascular phosphorylated tau as well as perivascular inflammation is observed with this type of CAA (see later).

The genetics of CAA is tightly linked to that of AD. However, research conducted on CAA has been more rewarding in improving our understanding of the biochemistry and molecular genetics of Aβ amyloidogenesis. Not only was Aβ peptide identified first from amyloid deposited in vascular walls in AD and Down’s syndrome patients [25], but the APP mutations were also identified for the first time in a familial form of CAA called hereditary cerebral haemorrhage with amyloidosis – Dutch type (HCHWA-D). HCHWA-D is an autosomal dominant form of CAA characterized by recurrent haemorrhages due to extensive Aβ deposition in cerebral blood vessel walls and diffuse amyloid plaques but absence of dense-core senile plaques or neurofibrillary tau pathology [26,27,28] (Table 1). HCHWA-D patients were identified to carry Dutch APP mutation at codon 693 (APP 770 isoform) replacing a glutamate (E) by a glutamine (Q) [29,30].

Soon after, the Flemish APP mutation replacing a glycine (G) by an alanine (A) at the adjacent codon 692 was identified in another Dutch family suffering from presenile dementia and cerebral haemorrhage inherited in an autosomal dominant pattern [31]. Interestingly, clinical phenotypes in Flemish APP carriers overlapped to the extent that cerebral haemorrhages were reported in offspring of patients having dementia while patients with strokes had offspring developing progressive dementia [32,33]. Flemish APP carriers were also shown to have definitive AD with numerous, large senile plaques accompanied by a very severe degree of leptomeningeal and parenchymal CAA [34,33,35]. Moreover, in contrast to HCHWA-D patients, Flemish APP carriers have neurofibrillary tau pathology [33,34] (Figure 2).

Subsequently, other APP mutations within the Aβ sequence were also identified – such as “Italian” (E693K), “Arctic” (E693G), “Japanese” (E693Δ), “Iowa” (D694N) and L705V APP mutations (Table 2; www.molgen.ua.ac.be/ADMutations). The pathology associated with these mutations resembled either Dutch or Flemish APP pathology by having a pure CAA/cerebral haemorrhage or a mixed AD – CAA/cerebral haemorrhage phenotype. For instance, Italian APP and APP L705V mutation resembles the Dutch APP carriers with a predominant CAA with or without recurrent haemorrhages [36,37]. Also, similar to Dutch APP mutation, neurofibrillary tangles (NFT) were absent in APP L705V carriers [37] or were mild and restricted to the archicortex as in the Italian APP carriers [36]. The Arctic and Iowa APP carriers resemble the Flemish APP carriers by having a mixed AD/severe CAA phenotype, although haemorrhagic strokes are not observed [38,39,40,41]. Moreover, consistent to an AD phenotype, parenchymal, ringlike plaques are observed in the Arctic APP carriers [39]. In addition to these APP mutations near the α-secretase site, intragenic Aβ mutations are also reported near the β- and γ-secretase site of APP. For instance, APP A713T mutation near the γ-secretase site reportedly leads to progressive dementia and multiple strokes, and pathological studies confirm AD with severe CAA with multiple infarcts in brain [42]. Similarly, the APP A673V mutation near the β-secretase site of APP has been shown to be associated with AD type of dementia and, consistent with an autosomal recessive pattern of inheritance, is pathogenic only in a homozygous state [43]. Neuropathological data for APP A673V carriers is not yet available [43].

It is clear that APP mutations lying within the Aβ sequence cause pathology by mechanisms distinct from those involving the other APP or presenilin (PS) mutations that cause AD. For instance, the AD-causing mutations in APP or PS were shown to alter APP processing in a manner that increases the absolute or relative levels of Aβ42, the more fibrillogenic Aβ species [47]. Consider, for example, APP Swedish double mutation (K670N/M671L) that leads to increased absolute levels of Aβ42 along with Aβ40 (without changing the Aβ42/Aβ40 ratio), while γ-secretase site APP mutations increase the relative levels of Aβ42 by increasing the ratio of Aβ42/Aβ40 [4]. Similarly, PS1 and PS2 mutations also increased the relative levels of Aβ42 by increasing the ratio Aβ42/Aβ40 [4]. On the other hand, mutations within the Aβ sequence were predicted not only to alter the processing of APP and/or increasing total Aβ production, but were also thought to alter the aggregation properties of the resulting mutant Aβ peptide.

Initial studies seemingly supporting this hypothesis, however, were not completely clear. This was because the Dutch APP peptide showed highly accelerated fibrillogenic kinetics, increased stability, and higher in vitro toxicity; while the kinetics of the Flemish Aβ peptide were even slower than the wild-type Aβ and the peptide also did not show any increased toxicity to cells in vitro [48,49,50,51]. This obviously contrasted sharply with the clinicopathological consequences as Flemish APP carriers had clinical and pathological AD while the Dutch APP patients although had diffuse type of Aβ deposits, but lacked neuritic, senile plaques characteristic of AD. Subsequent in vitro studies, however, showed that compared to wild type and Dutch Aβ, Flemish Aβ was most toxic to differentiated SHSY5Y cells in the early stages of aggregation when Aβ fibrils were not observed [51]. In the late stages of aggregation, the Dutch peptide remained the most toxic Aβ species, suggesting that Aβ, at least under in vitro conditions, is neurotoxic in an initial phase due to its soluble oligomeric or other early toxic Aβ intermediate(s), which was distinct from the late neurotoxicity incurred by larger aggregated assemblies of Aβ [51,52]. Moreover, it was also suggested that the cytotoxic potential of Aβ lies in their ability to form extensive fibrils directly on the cell surface, as preaggregation of Aβ abolished its toxic effect on cultured cells [53]. Thus, Dutch Aβ once released from the sites of generation, aggregates almost instantaneously in the parenchymal matrix as non-neuritic diffuse plaques [51,54]. The Flemish APP, on the other hand, not only leads to increased Aβ production due to an increased activity of β-site APP-cleaving enzyme-2 [55], but also slowly fibrillizing mutant Aβ diffuses more efficiently to aggregate as one of the largest dense-core amyloid deposits known in AD. Subsequently, it was also shown that most of the dense-core amyloid plaques observed in the Flemish APP carriers develop in close association with vessel walls [33], perhaps due to entrapment of Aβ in the vascular clearance routes [56,57]. Further observations in Flemish APP carriers showed that CAA-related plaques could also trigger accumulation of tau-immunoreactive dystrophic neurites in the surrounding neuropil similar to dense-core plaques [33] (Figure 2). Vessel association of dense-core plaques might not be unique to Flemish APP carriers as suggested by a recent study showing a significant association between dense-core plaques and Prussian blue-labelled haeme deposits in sporadic AD and Down’s syndrome patients, and even proposing that dense-core plaques are sites of older microhaemorrhages [58].

Subsequent studies on other APP mutations within the Aβ sequence identified similar or additional disease mechanisms. Consider for example, the Italian peptide showing increased propensity to make fibrils and peptide-mediated pathogenic effects similar to the Dutch peptide [59,60]. As a matter of fact, substitutions at APP 693 codon that cause either a loss of charge (E22Q; Dutch APP) or a change of charge (E22K; Italian APP) show increased binding to and degeneration of cerebrovascular smooth muscle cells [59] and, not surprisingly, both of them lead to CAA/haemorrhagic strokes. On the other hand, complete deletion of APP 693 codon as in Japanese APP shows altered aggregation property of enhanced oligomerization but no fibrillization [45]. This is similar to the Arctic APP variant that also increases the propensity of protofibril formation without increasing the rate of fibril formation or the production of Aβ [38]. And lastly, the Iowa APP carriers that resemble Flemish and Dutch APP carriers show cytotoxicity and aggregation properties that lie between those caused by the Flemish and Dutch Aβ peptides [60]. Interestingly, Flemish, Dutch, Italian, Arctic, and Japanese Aβ variants are also shown to be more resistant to proteolytic degradation [61,45,62]. Recent data also suggest that Dutch, Italian, and Iowa preferably assemble in the presence of GM3 ganglioside [63] while their clearance across the blood-brain barrier (BBB) might be reduced as shown for Dutch/Iowa mutant [64,65]. It remains to be shown that oligomers of the variant Aβ that are associated with AD changes are more synaptotoxic than wild type Aβ oligomers [45].

Copy number variations of the APP gene also seem to be important for development of CAA. For instance, Down’s syndrome patients who have three copies of APP, have CAA as young as 30 years and the severity of CAA increases with age [66]. Down’s syndrome patients are also reported to have strokes [67,68,69]. Similarly, duplications of the APP locus in certain French and Dutch families or populations demonstrate an autosomal dominant AD and/or lobar cerebral haemorrhage with prominent CAA [70,71]. APP duplications increase the absolute amounts of both Aβ42 and Aβ40 (without changing the ratio), an important factor proposed to lead to Aβ deposition in vessel walls [33,72]. Although uncommon, a predominant CAA is also observed in select PS1 mutations like Q184D and L282V, amongst others [73,74,75,76,77]. The best studied PS2 mutation, N141I, has also been reported to lead to haemorrhagic strokes in Volga-German family [78]. It remains to be shown whether some of these mutations, especially those occurring after codon 200 [79], might not alter Aβ40 production as has been shown for the majority of PS mutations [80,81].

Specific apolipoprotein E (APOE) alleles are not only a strong risk factor for development of AD, but are also linked to development of CAA and strokes. CAA related to capillaries and smaller arterioles is observed to be closely associated with ApoE-ε4, but not ApoE-ε2 [7]. ApoE-ε4 is also independently associated with increased vascular Aβ deposition in large-vessel CAA leading to strokes [82,83,84]. On the other hand, ApoE-ε2 appears to promote degenerative changes in the amyloid-laden vessel wall and is a cause of stroke independent of Aβ deposition [85,86]. Anecdotal examples of individuals homozygous for the APOE-ε4 allele that have a predominant CAA or a CAA-related pathology are also described [23]. Lastly, candidate genes involving Aβ degradation pathways have also been implicated in the development of CAA [87]. For instance, a polymorphism in the Neprilysin gene has been shown to be associated with CAA [88]. For many of these candidate genes, knockout mouse models have been made as discussed in the next section.

Most disease models are based on inherited forms of disease that although less common than the sporadic forms, are clinically indistinguishable from each other. The first attempts to model the archetypal Dutch and Flemish APP cerebral amyloidosis in mice did not show amyloid till at least 18 months of age [89,90], however, recent experiments have successfully created Dutch APP mice [91]. At ≈ 25 months of age, APP Dutch mice demonstrate extensive CAA, smooth muscle cell degeneration and haemorrhages – features that are typical of HCHWA-D [91]. Similarly, transgenic mice expressing Dutch/Iowa mutant Aβ (Tg-SwDI) at levels below those of endogenous mouse APP have also been established that robustly deposit Aβ, particularly in the cerebral microvasculature, from 3 months of age [92,93]. However, because mutant Aβ is expected to have altered biophysical properties, any direct extrapolation to sporadic CAA has to be made with caution.

Interestingly, mice overexpressing wild type Aβ such as APP Swedish mice (APP/Sw; e.g., Tg2576 and APP23 mice), or APP London mice, develop CAA fairly early and have been used as a valid model for CAA [94,95,96](www.alzforum.org/res/com/tra). Importantly, APP/Sw mice with unaltered relative ratios of Aβ42/Aβ40 are even more representative of sporadic CAA/AD and have been employed in several studies. In principle, these mouse models have provided support for many important concepts in our understanding of cerebral amyloidosis including Aβ deposition in vessel walls, as summarized below:

Mouse models have given a definitive view that Aβ42 is essential for Aβ deposition not only in parenchyma but also in vessels. When APP/Sw mice (producing both Aβ40 and Aβ42) are bred with mice expressing mutant PS1 (that increases brain Aβ42/Aβ40 ratio), the crossbred mice have drastically increased amyloid deposition in parenchyma and vessels compared to single transgenic controls [97,98]. These APP/PS1 mice also have a higher number but smaller sizes of amyloid deposits most likely due to the extra “seeds” provided by Aβ42 [99]. A similar crossbreeding of APP Dutch mice with mice expressing AD-related PS1 G384A mutation also increases amyloid depositions and shifts the pathology from vascular to the parenchymal compartment [91]. The premise that Aβ42 is essential for Aβ deposition is also neatly answered by BRI-Aβ40 and BRI-Aβ42 mice that produce only Aβ40 or Aβ42 [100]. In these mice, a fusion construct is utilized wherein the carboxyl terminus sequence of BRI protein, involved in amyloid deposition in familial British and Danish dementia, is replaced by a sequence encoding either Aβ40 or Aβ42. A proteolytic cleavage of this fusion protein at a furin cleavage site immediately preceding Aβ results in high-levels of Aβ40 or Aβ42 secretion. While BRI-Aβ40 mice expressing high levels of Aβ40 do not develop overt amyloid pathology, the BRI-Aβ42 mice line expressing lower levels of Aβ42 develop all types of brain amyloid deposits including CAA [100]. Crossbreeding of BRI-Aβ42 mice with Tg2576 mice again leads to a massive increase in amyloid deposition. These data establish that Aβ42 is essential for amyloid deposition in the parenchyma and also in vessels.

In this review, I have discussed how molecular pathology of hereditary cerebrovascular amyloidosis due to APP mutations within the Aβ domain, as well as experimental studies in transgenic mouse models have helped to partially dissect molecular mechanisms involved in Aβ cerebrovascular amyloidosis. Firstly, the mutant forms of Aβ causing AD/cerebral haemorrhage were discussed that strongly support the role of APP in the disease mechanism and provide one of the strongest rationales for studies of factors that influence abnormal metabolism and aggregation of Aβ in the causation of CAA with or without AD. Secondly, studies on hereditary cerebrovascular amyloidosis and mouse models were discussed that suggest that while an absolute or relative increase in Aβ42 or an increased Aβ42/Aβ40 ratio is important for parenchymal plaque deposition and development of an AD phenotype, increased Aβ with an unaltered Aβ42/Aβ40 ratio favours cerebrovascular amyloidosis. However, critical relative levels of Aβ42 are also important as without this, the clearance of Aβ is too efficient to allow deposition. Thirdly, these data suggest that a low fibrillogenic potential of Aβ could also favour plaque compaction as is the case with Flemish Aβ that leads to the formation of large, neuritic, dense-core plaques. And lastly, by illuminating the relationships between specific lesions such as CAA and dense-core plaques and their molecular components, studies on hereditary cerebrovascular amyloidosis and mouse models have shown that CAA and dense-core plaques could be a spectrum of the same disease mechanism. Thus, molecular pathological studies of cerebrovascular amyloidosis and relevant mouse models would continue to provide detailed insights into the pathogenesis of CAA and contribute to the development of targeted therapeutic strategies.