Emerging concepts in sporadic cerebral amyloid angiopathy

Sporadic cerebral amyloid angiopathy is a common, well-deﬁned small vessel disease and a largely untreatable cause of intracerebral haemorrhage and contributor to age-related cognitive decline. The term ‘cerebral amyloid angiopathy’ now encompasses not only a speciﬁc cerebrovascular pathological ﬁnding, but also different clinical syndromes (both acute and progressive), brain parenchymal lesions seen on neuroimaging and a set of diagnostic criteria—the Boston criteria, which have resulted in increasingly detected disease during life. Over the past few years, it has become clear that, at the pathophysiological level, cerebral amyloid angiopathy appears to be in part a protein elimination failure angiopathy and that this dysfunction is a feed-forward process, which potentially leads to worsening vascular amyloid- b accumulation, activation of vascular injury pathways and impaired vascular physiology. From a clinical standpoint, cerebral amyloid angiopathy is characterized by individual focal lesions (microbleeds, cortical superﬁcial siderosis, microinfarcts) and large-scale alterations (white matter hyperintensities, structural connectivity, cortical thickness), both cortical and subcortical. This review provides an interdisciplinary critical outlook on various emerging and changing concepts in the ﬁeld, illustrating mechanisms associated with amyloid cerebrovascular pathology and neurological dysfunction. phenotypes of CAA, clinically expressed with more cognitive impairment and probably more strongly associated with cortical microinfarcts with or without Alzheimer’s disease. intermediate phenotypes of the disease. genotype and vascular risk factors all the critical and different directions during the course of the ageing


Introduction
Cerebrovascular deposition of amyloid-b (also known as cerebral amyloid angiopathy, CAA) is a strikingly common finding in elderly people and a major cause of spontaneous intracerebral haemorrhage (ICH), as well as an important contributor to age-related cognitive decline (Viswanathan and Greenberg, 2011;Boulouis et al., 2016). However, it is not a long time since CAA was considered a rather obscure pathological curiosity and innocent bystander in the elderly brain, often accompanying Alzheimer's disease. Despite the close molecular relationship between the two entities, CAA remains clinically distinct from Alzheimer disease, but has the potential to link cerebrovascular and neurodegenerative pathways in the ageing brain (Cordonnier and van der Flier, 2011). The CAA story is remarkable in that unlike most neurological disorders, the recognition of the underlying pathology preceded by many years the association with clinical syndromes and disease. In 1954, Stefanos Pantelakis from the University Psychiatric Clinic in Bel-Air, Switzerland, highlighted CAA as a distinct entity (Pantelakis, 1954). In his landmark paper, Pantelakis raised the hypothesis that vascular amyloid may originate in the brain (in line with modern concepts of amyloid pathophysiology) andamong other things-described what are still the pathological hallmarks of CAA: (i) preferential involvement of small arterioles and capillaries of the leptomeninges and cerebral cortex, without necessarily involving adjacent parenchymal; (ii) topographical distribution favouring posterior lobar brain regions (especially the occipital lobes); (iii) lack of involvement of white matter small vessels; (iv) association with increased age and dementia; (v) lack of association with hypertension and arteriosclerosis (the other main small vessel disease, which preferentially affects the basal ganglia and brainstem); and (vi) lack of any link with systemic amyloidosis.
Since these early descriptions, understanding of the pathophysiological mechanisms and clinical imaging manifestations of CAA has increased substantially. CAA denotes not only a specific cerebrovascular pathological trait, but also a clinical syndrome (or syndromes), and a set of clinical imaging diagnostic criteria (the Boston criteria), which have resulted in increasingly detected disease during life (Linn et al., 2010;Charidimou et al., 2012a). Over the past decade we have also come to realise that the integrity of cerebral microvessels is critical for solute efflux from the interstitial fluid of the brain, as these molecules pass from brain into the perivascular drainage system and the systemic lymphatic circulation (Carare et al., 2013). Cerebrovascular accumulation of amyloid can thus be viewed in some ways 'as a canary in a coal mine', an indicator of perivascular drainage failure (Carare et al., 2013;Keable et al., 2016). This process may set in motion a damaging disease process that both further exacerbates the clearance defects and triggers clinically relevant haemorrhagic (Zhao et al., 2015) and ischaemic brain injury .
This review update focuses on selected recent topics that have not been extensively covered in previous reviews and illustrate emerging and changing concepts in sporadic CAA, including the neurological effects and underlying biology of the disease. Given the fast-moving CAA research landscape (Charidimou et al., 2016a) in the current paper we are not covering all the clinical and imaging aspects of CAA in detail, given the more specific published reviews (Greenberg et al., 2009b(Greenberg et al., , 2014Pantoni, 2010;Smith et al., 2012b;Reijmer et al., 2016b). Instead, emphasis is given to concepts, findings and theoretical frameworks of the disease, with examples from the literature to illustrate these points. Finally, our paper focusses on the more common sporadic CAA-for the rare genetic forms of cerebrovascular amyloid accumulation and CAA-related inflammation, please see other reviews and recent publications for further information (Revesz et al., 2009;Charidimou et al., 2012a;Auriel et al., 2016).

Cerebral amyloid angiopathy clinical aspects
Pathologically-defined CAA is common in the elderly and advancing age is the strongest known risk factor for sporadic CAA (Vinters, 1987). Population-based autopsy studies indicate a CAA prevalence of 20-40% in nondemented, and 50-60% in demented elderly populations (age range for both groups 80-90 years) (Keage et al., 2009). For severe CAA, prevalence in demented and nondemented elderly groups is 30-40% and 7-24%, respectively (Keage et al., 2009). In Alzheimer's disease brains, CAA is identified in an estimated 85-95% of the cases when sought diligently (Kalaria and Ballard, 1999;Jellinger, 2002). Only a relatively small proportion of these older individuals are diagnosed during life with CAA-related clinical symptomatology (acute or progressive). The sentinel clinical presentations of CAA include spontaneous lobar ICH, cognitive impairment and dementia (either post ICH or in patients without stroke) and transient focal neurological episodes ('amyloid spells') often associated with acute convexity subarachnoid haemorrhage or cortical superficial siderosis. The relative frequency of the different CAA clinical presentations is hard to assess, as it is heavily dependent on making an accurate diagnosis of underlying CAA in different settings using appropriate MRI sequences (including blood sensitive sequences). The association of these presentations with objective assessments of clinical frailty, comorbidities or chronological age per se are under investigation. The key neuroimaging signatures of the disease on clinical MRI include multiple strictly lobar cerebral microbleeds, cortical superficial siderosis, white matter hyperintensities, cortical microinfarcts and MRI-visible perivascular spaces in the centrum semiovale (Greenberg et al., 2014). An overview of the key CAA clinical presentations and the significance/role of different MRI markers of small vessel disease in each of these based on the totality of current evidence is provided in Table 1.

Dementia and cognitive impairment in cerebral amyloid angiopathy
Cognitive impairment in CAA was first reported over two decades ago (Gray et al., 1985;Greenberg et al., 1993). Dissecting the independent impact of CAA on cognition, however, is confounded by co-existing Alzheimer's disease and other age-related pathologies (e.g. sporadic non-amyloid microangiopathy). Recent studies in older community dwelling persons have suggested an association between moderate-to-severe CAA pathology and lower perceptual speed and episodic memory (Arvanitakis et al., 2011). In these cohorts, CAA is a common finding and appears to independently contribute to the level of cognitive deficit beyond Alzheimer's disease or other accompanying pathologies (Pfeifer et al., 2002;Boyle et al., 2015). In one community-based cohort, individuals harbouring CAA not only had increased risk of dementia [odds ratio (OR): 1.237; 95% confidence interval (CI): 1.082-1.414] but a faster rate of decline in both global and domain-specific cognition (Boyle et al., 2015). In a hospital-based setting, patients with CAA without dementia evaluated for stroke or stroke-like symptoms appeared to be at high risk for developing dementia (Moulin et al., 2016). Interestingly, in this study, the history of previous CAA-related ICH was not associated with the degree of cognitive deficits. Indeed previous work has suggested that cognitive decline in CAA may precede ICH (Viswanathan et al., 2008;Cordonnier et al., 2010;Xiong et al., 2016b), thus indicating that CAA-related cognitive decline may occur independently of symptomatic ICH. In another hospital-based study, patients with CAA without dementia evaluated for stroke or stroke-like symptoms appeared to perform significantly worse on nearly all tests in a standard neurocognitive battery compared to a group of healthy controls with similar age and education level (Xiong et al., 2016a). In this study, these deficits were clearly present even though the majority of patients did not have self-reported cognitive concerns or cognitive deficits (Xiong et al., 2016a). In the analyses of the relationship between these cognitive deficits and neuroimaging biomarkers, lower total brain volume was associated with slower processing speed and worse executive function, and white matter hyperintensity volume was related to executive function in CAA patients (Xiong et al., 2016a).
Despite the clinical observation of cognitive impairment in CAA, the frequency, severity, and specific cognitive profile of CAA patients are not well understood (Schrag and Kirshner, 2016). To address this question, a pilot study reported neuropsychological and neuroimaging data from an ongoing prospective cohort (Functional Assessment of Vascular Reactivity in CAA) (Case et al., 2016). This study compared the neuropsychological test results in non-demented CAA participants (n = 34) to participants with dementia caused by Alzheimer's disease (n = 16), mild cognitive impairment (n = 69), and ischaemic stroke (n = 27) (matched for stroke severity to CAA participants). The mean test scores for CAA patients were significantly lower than norms for memory, executive function, and processing speed. A large proportion of these CAA patients (79%) fulfilled criteria for mild cognitive impairment based on low cognitive performance and cognitive concerns. Compared to Alzheimer's disease, CAA patients had similarly low executive function scores, but relatively preserved memory. CAA participants' scores were lower than those of the ischaemic stroke group for executive function and processing speed. Lower processing speed scores in CAA were associated with higher white matter hyperintensity volumes (Case et al., 2016).
Taken together, current evidence suggests that mild cognitive impairment and dementia are very prevalent in clinically diagnosed CAA patients. The overall cognitive profile of CAA is more similar to that seen in classic vascular cognitive impairment than Alzheimer's disease (i.e. executive dysfunction and impaired processing speed with relatively preserved episodic memory). White matter ischaemic lesions may underlie some of the impaired processing speed seen in CAA. CAA-related dementia and vascular cognitive impairment are likely the result of both small, spatially distributed brain injuries including cerebral microbleeds (Moulin et al., 2016), cortical microinfarcts, and altered structural connectivity. In support of this hypothesis, a recent study examining brain networks in CAA as a surrogate measure of global small-vessel pathology showed that lower global network efficiency was independently related to worse performance on tests of processing speed and executive functioning .
Clinical imaging expression and spectrum of cerebral amyloid angiopathy CAA is most often recognized in life by symptomatic, spontaneous ICH, preferentially affecting cortical-subcortical (lobar) regions (especially the occipital and posterior temporal lobes) (Rosand et al., 2005). These individuals have predominantly severe CAA (Alonzo et al., 1998), with small burden of neurofibrillary tangles and senile plaques on neuropathology (Charidimou et al., 2015c). Multiple recurrent CAA-related haematomas can occur over months or years (at a rate of recurrence on the order of 10% per year) (Biffi et al., 2010), with progressive neurological decline. CAA may also be an important risk factor or cause for ICH related to oral anticoagulation use and thrombolysis-related haematomas in acute ischaemic stroke (Charidimou et al., 2015d;Mattila et al., 2015). Reliable assessment of an individual's haemorrhagic burden over time and prevention of future ICH is thus a key component in the clinical management of CAA patients (see below).
In this context, characterization of the expanding clinical imaging spectrum of CAA has been refined substantially in the past years with the broader availability of brain MRI . In addition to lobar ICH, other characteristic MRI biomarkers of CAA now include strictly lobar microbleeds, cortical superficial siderosis, white matter hyperintensities (especially posteriorpredominant), MRI-visible perivascular spaces in the cerebral white matter and cortical microinfarcts (Fig. 1). These neuroimaging markers probably reflect related but distinct aspects of CAA pathophysiology (Charidimou and Jager, 2014). An important advance in the field has been the recent publication of international consensus standards for describing and reporting many of these small vessel disease lesion types (Wardlaw et al., 2013). Multiple strictly lobar microbleeds detected on T 2 * /SWI (susceptibility-weighted imaging) MRI are accepted as one of the hallmark biomarkers for CAA presence and are useful for CAA diagnosis within the Boston criteria (Knudsen et al., Figure 1 Schematic representation of the spectrum of haemorrhagic and ischaemic manifestations of sporadic CAA, visible on structural MRI, including three key common neuropathological lesions seen in CAA brains. (A) A lobar cerebral microbleed identified on post-mortem 7 T MRI in the brain tissue of an 81-year-old male with dementia and severe CAA on pathology. On haematoxylin and eosin stain, brown and yellow deposits representing haemosiderin, and haematoidin, are seen, indicating that this haemorrhage is not chronic but subacute. (B) Prussian blue stain from an area corresponding to cortical superficial siderosis on in vivo MRI of a patient with advanced CAA. Blue deposits corresponding to haemosiderin are seen in the subarachnoid space (SAS) surrounding a leptomeningeal arteriole (which shows a vessel-within-vessel appearance) and in the superficial layers of the cortex. (C) A cortical microinfarct identified on histopathological examination. Note the severe vascular amyloid deposition (dark brown: immunostained for amyloid-b) in both the overlying leptomeningeal small vessels of different diameters and two cortical arterioles in the vicinity of the microinfarct. Bottom: Imaging CAA in clinical practice. 1.5 T brain MRI sequences. (D) Seventy-fiveyear-old male; initial evaluation of stereotyped episodes of right sided of tingling in the past year, axial section of SWI demonstrating a focus of cortical superficial siderosis on both banks of the left central sulcus, and multiple strictly lobar microbleeds (arrowheads in magnification). (E) Eighty-eight-year-old female with history of cognitive complaints, 6 months after initial imaging demonstrating multiple lobar microbleeds. Persistent headache and recent episode of left hand numbness and weakness; axial section of FLAIR sequence showing a linear hyperintensity in the right central sulcus, corresponding to acute subarachnoid blood. (F) Seventy-seven-year-old male otherwise meeting Boston criteria for CAA; coronal section (top) of T 1 -weighted, and axial section of T 2 -weighted sequences showing multiple dilated perivascular spaces (magnification) in the centrum semiovale, following the path of small calibre arteries. Diffuse widening of cerebral sulci due to marked cortical atrophy can also be appreciated. (G) Sixty-five-year-old male with acute onset of right hemiplegia and impaired consciousness at hospital arrival. Large right lobar (mostly parietal) ICH ruptured into the convexities and subarachnoid space. Contralateral multiple strictly lobar microbleeds are seen. EPVS = enlarged perivascular space; CMBs = cerebral microbleeds; sSAH = spinal subarachnoid haemorrhage; WMH = white matter hyperintensity. Linn et al., 2010) and to assess disease evolution (Pasquini et al., 2016). Strictly lobar cerebral microbleeds within these criteria have a high positive predictive value for CAA even in individuals without ICH presentations in a hospital-based setting (specificity 490% and positive predictive value 487%) . This is of particular clinical relevance, as CAA is increasingly recognized to present without major haemorrhage. Lobar microbleeds also frequently occur among presumably healthy people as a result of previously clinically silent CAA (Mesker et al., 2011), but also in the absence of CAA, yielding low overall specificity in this setting . However, the detection of cerebral microbleeds in different clinical settings raises thorny clinical dilemmas regarding their biological significance and how they might affect treatment decisions, especially regarding antithrombotic drug use (Greenberg et al., 2009b;Wang et al., 2014). Clinical radiological criteria for CAA-related inflammation have also been validated , allowing non-invasive diagnosis of this potentially treatable meningoencephalitis syndrome .

2001;
Recent data have implicated acute or chronic haemorrhage within or adjacent to the cortical sulci (often described as cortical superficial siderosis when chronic or as sulcal subarachnoid haemorrhage when acute) as key haemorrhagic neuroimaging signatures of CAA (Charidimou et al., 2015b). They likely reflect repeated episodes of blood leaking into the subarachnoid space from brittle and fragile CAA affected vessels. Siderosis is emerging as perhaps the most clinically relevant manifestation of the disease: (i) it is a trigger of transient focal neurological symptoms ('amyloid spells') (Greenberg et al., 1993;Charidimou et al., 2012b), expanding the CAA clinical imaging spectrum (Raposo et al., 2011;Calviere et al., 2016); (ii) it has been shown to carry a very high risk of future symptomatic lobar ICH (Linn et al., 2010;Charidimou et al., 2013c) with important implications for CAA clinical care, such as strict blood pressure control (Gorelick et al., 2011;Biffi et al., 2015;Hemphill et al., 2015;Carcel et al., 2016) and avoiding antithrombotics unless otherwise strongly indicated (Biffi et al., 2010;Charidimou et al., 2012a); and (iii) it may be an independent risk factor for new onset dementia after ICH (Moulin et al., 2016). Of note, siderosis exerts these effects independent of, and above any effect of coincident microbleeds Biffi et al., 2010). Recently, consensus standards for rating and reporting cortical superficial siderosis have been suggested and are encouraged for use in observational and clinical studies (Charidimou et al., 2015b). The emerging key role of cortical superficial siderosis in neurological dysfunction in CAA might be linked with the neuropathological observation of more severe leptomeningeal CAA compared to parenchymal and between superficial and deep cortical layers in any given brain (Kovari et al., 2013).

Role of molecular imaging in cerebral amyloid angiopathy
While CAA appears detectable by molecular amyloid imaging (Klunk et al., 2004;Bacskai et al., 2007;Greenberg et al., 2008) the clinical diagnostic utility of amyloid PET in suspected CAA remains undefined. Non-demented patients with CAA show higher global Pittsburgh compound B (PiB)-PET retention when compared to healthy control subjects and higher occipital-to-global PiB ratios compared to patients with Alzheimer's disease Ly et al., 2010). However, due to the frequent occurrence of high PiB-PET uptake in the healthy elderly likely reflecting incipient Alzheimer disease, specificity is limited; one study reported sensitivity of 91%, and specificity of 55% compared to healthy controls . A negative PiB scan appears to rule out severe CAA and may in the future have clinical research implications for prognostication, treatment decisions, and patient selection for drug trials. A small study recently tested the clinical relevance of florbetapir-PET, another amyloid tracer, in distinguishing CAA-related lobar ICH from hypertensive ICH . In a study sample of cognitively normal patients, including 10 lobar haemorrhages with a clinical imaging diagnosis of probable CAA and nine with deep haemorrhages and without evidence for CAA of similar age (66.9 versus 67.1 years), sex, and leukoaraiosis volumes, the mean global cortical florbetapir uptake was higher in CAA versus non-CAA (P = 0.001). Based on visual rating for positive/negative florbetapir, all 10 patients with CAA versus 1/9 non-CAA patients were classified as positive (sensitivity: 100%; 95% CI: 66%-100% and specificity: 89%; 95% CI: 51-99% for determination of probable CAA) . The general conclusion is similar to previous studies using PiB-PET: a negative scan almost certainly rules out probable CAA-at least in these specific and well-defined comparison groups used in the study, 'probable' CAA with haemorrhage and deep hypertensive-related haemorrhage.
The clinical usefulness of amyloid-PET for CAA diagnosis in everyday practise currently remains limited and under investigation. Further larger prospective studies are needed to provide external validation, and test whether amyloid-PET can be useful for detecting CAA pathology in cases where the diagnosis is uncertain. Examples would include cases with mixed location of ICH (both lobar and deep), single lobar haemorrhage without microbleeds (labelled as 'possible' CAA) as well as cases with microbleeds or cortical superficial siderosis without any haematomas. A large prospective study of such uncertain cases with long followup is required to determine the sensitivity and specificity of amyloid imaging in these settings and at earlier stages of suspected CAA. Development of a molecular imaging tracer specific for vascular amyloid would be a major step forward for the field and preclinical studies continue to explore possible candidates (Jia et al., 2014(Jia et al., , 2015. Other novel methods, such as early-phase PiB-PET (Farid et al., 2015) and tau imaging, might help test similarities and differences between CAA and Alzheimer's disease in living individuals (Villemagne et al., 2015), although given the strong overlap between these pathologies, it is unlikely that any technique in isolation will fully distinguish these two entities.
Amyloid imaging has recently become an important tool in CAA research to explore mechanisms and test spatial and quantitative correlations between proposed CAArelated brain lesions and the inciting pathology (i.e. vascular amyloid). In a cross-sectional study, higher PiB retention was detected at the sites (i.e. shells concentrically surrounding the bleeds) of lobar microbleeds in CAA when compared to other locations with high-probability to show CAA-related microbleeds (Dierksen et al., 2010). Using a similar approach, in a longitudinal analysis of follow-up MRIs (median of 19 months after baseline), sites of future microbleeds demonstrated increased PiB retention relative to simulated lesions placed according to a probability density map (distribution volume ratio 1.34 versus 1.14, P 5 0.0001) with fall-off at increasing distances from sites of future microbleeding (Gurol et al., 2012). Despite the limitations imposed by the relatively poor spatial resolution of amyloid-PET, resulting in some imprecision for co-localizing PiB retention and haemorrhage location, these studies provide some proof of the concept that vascular amyloid deposition might be a key contributor to lobar microbleeds and macrobleeds. However, the exact mechanisms for these associations remain uncertain. Also, another study showed a positive correlation between leukoaraiosis volume and global PiB retention in patients with CAA but not in healthy controls or patients with Alzheimer's disease, suggesting that predominanlty vascular amyloid deposition rather than parenchymal might drive this association (Gurol et al., 2013). More recently, a study suggested that MRI-visible high degree centrum semiovale perivascular spaces, a novel CAA-related MRI marker, were correlated with a higher median amyloid PiB retention across a wide range of cerebrovascular amyloid deposition, after adjusting for age, microbleeds and white matter hyperintensities (Charidimou et al., 2015a).

Cerebral amyloid angiopathy: the search for clinical trials surrogate markers
Although conventional CAA neuroimaging biomarkers are currently used for diagnosis and clinical monitoring, none has yet emerged as a valid clinical trials surrogate marker (Greenberg et al., 2014). This apparent paradox might be partly due to the fact that these haemorrhagic and nonhaemorrhagic structural focal lesions used clinically are only the tip of the iceberg and fail to capture the cumulative large-scale impact of the disease. Also, the MRI-histopathological basis of small vessel disease neuroimaging biomarkers still needs to be defined, representing a remarkable lacune in the field (Gouw et al., 2011). For example, lesion-by-lesion analysis has been reported for fewer than 100 microbleeds (Fazekas et al., 1999;Tatsumi et al., 2008;Schrag et al., 2010) often using relatively insensitive T 2 * -weighted techniques. Of note, the large numbers of cerebral microbleeds noted by MRI are rarely seen in autopsy brain specimens, a disconnect that likely reflects the ability of T 2 * -weighted MRI to fully sample the brain to an extent not achievable by histology. Although microbleeds are strongly associated with advanced CAA and it is commonly assumed that amyloid deposition in the walls of cortical vessels directly causes microbleeding, a recent detailed ex vivo 7 T MRI study challenged this notion. In 7/19 identified microbleeds, the presumed involved vessel could be identified on histopathological sectioning. Only one of these vessels was positive for amyloid-b at the site of rupture. Moreover, the density of amyloid-b positive cortical vessels was lower in areas surrounding microbleeds, compared to control areas (van Veluw et al., 2017). Extremely limited data exist for perivascular spaces and siderosis. Brain banks (comprised of large brain donation programmes and collaborative initiatives) will be essential, in combination with new research techniques, in validating biomarkers in the field (Samarasekera et al., 2013;McAleese et al., 2016;Skrobot et al., 2016).
For CAA-related cognitive impairment, a major reason that no single neuroimaging surrogate marker has emerged is the mismatch between the lesion types that are most MRI detectable versus those that are actually numerous enough to substantially damage brain connections and mediate cognitive dysfunction. Cerebral microbleeds or lacunes of presumed vascular origin (Wardlaw et al., 2013) for example, are readily visualized by MRI, but their total burden is typically only one to two per brain (Longstreth et al., 1998;Vernooij et al., 2007Vernooij et al., , 2008 in community dwelling elderly, and 510-20 even in brains with advanced small vessel disease Viswanathan et al., 2006a;Herve et al., 2009). Cerebral microinfarcts hypothesized to play a key role in vascular cognitive impairment (Smith et al., 2012b), conversely, are estimated to number in the hundreds or thousands per CAA brain (Westover et al., 2013;Auriel et al., 2015), but at mean diameter of $200 mm (Okamoto et al., 2009;Smith et al., 2012a;Westover et al., 2013), are largely invisible to conventional MRI.
Finally, haemorrhagic and non-haemorrhagic structural lesions discussed seem to represent vascular-mediated injury to the brain rather than abnormalities of the vessels themselves. Thus, they probably measure late and irreversible steps in the postulated pathways leading from vascular dysfunction to brain injury and clinical dysfunction.
Neuropathological aspects of cerebral amyloid angiopathy and differences with 'hypertensive arteriopathy' CAA results from a chronic degenerative process by which the media of parenchymal arterioles undergoes progressive loss of its smooth muscle cells with simultaneous accumulation of an eosinophilic hyaline material (Attems et al., 2011), mostly composed of the more soluble, amyloid-b 40 species. This is in contrast to amyloid plaques found in Alzheimer's disease, predominantly composed of amyloid-b 42 species. The vessels affected by amyloid-b can show secondary vasculopathic changes, including fibrinoid necrosis, loss of smooth muscle cells, wall thickening, microaneurysm formation, and perivascular blood breakdown products deposition (Love et al., 2014). A specific and important subtype of CAA is capillary CAA, which has received lots of attention recently (Attems et al., 2011). This forms the basis for the CAA classification into two pathological types: CAA type 1, characterized by amyloid in cortical capillaries (with or without involvement of other vessels), and CAA type 2, where amyloid deposits are restricted to leptomeningeal and cortical arteries, but not capillaries (Attems et al., 2011). Recently, the neuropathological and clinical characteristics of capillary CAA were investigated for the first time in a subset (n = 300; aged 5 85 years) of the Vantaa 85 prospective population-based study (Makela et al., 2015). The authors reported that capillary CAA was most frequent in the occipital lobe, while CAA type1 was associated with the severity of CAA overall (P 5 0.001), dementia (P 5 0.001), severe Alzheimer's-type neuropathology (P-value 0.09 for CERAD C and 0.017 for Braak stages V-VI), and APOE "4 (but not "2) allele carrier status (P 5 0.001) (Makela et al., 2015). Immunohistochemical analysis in one study revealed that capillary CAA is associated with altered expression of sphingolipids, likely reflecting inflammatory pathways involvement (de Wit et al., 2016).
When CAA is noted in the subarachnoid space, the amyloid deposits are usually adventitial rather than medial in the walls of larger affected arterioles in a chunky, spirallike fashion, suggesting they may have resulted from aggregates of amyloid-b that came to lodge in arterial adventitia through the perivascular drainage system (Vinters, 2015).
The distribution of CAA pathology shows a characteristic patchy pattern (Vinters, 1987), whereby foci of vessels severely affected by CAA may be adjacent to other vessel segments with mild or absent amyloid-b deposition (Vinters, 1987;Attems et al., 2011), contributing to the heterogeneity of the disease. The severity and extent of histological changes provide the basis of several CAA neuropathological scoring systems (Vonsattel et al., 1991b;Olichney et al., 1995;Thal et al., 2003), each with strengths and limitations (Attems, 2005;Attems et al., 2011). Recently, a standardized consensus neuropathological protocol for rating CAA was developed (Love et al., 2014), potentially allowing comparison of pathological data across centres.
In contrast to CAA, which is relatively easy to define, the other common subtype of age-related small vessel disease remains less clearly defined at a diagnostic or molecular level and is broadly described as 'hypertensive arteriopathy' (Pantoni, 2010). The umbrella term 'hypertensive arteriopathy' groups together a spectrum of sporadic non-amyloid small vessel disease pathologies often associated with advanced age (but not clearly age-driven), hypertension (though the disease is not necessarily, or even often, related specifically to hypertension), diabetes mellitus, and other common vascular risk factors (Pantoni, 2010). Hypertensive arteriopathy is characterized pathologically by collagenous thickening of the vessel wall with narrowing of the lumen and progressive loss of smooth muscle, and sometimes by exudation of fibrin and other serum proteins or by scanty mural deposition of lipid (Pantoni, 2010). Different pathological subtypes include arteriolosclerosis, fibrinoid necrosis, and lipohyalinosis (Pantoni, 2010;Charidimou et al., 2016b). The term 'sporadic non-amyloid microangiopathy' has been recently suggested as a more accurate alternative to group together these diseases .
Despite the limitations in definitions and the lack of widely accepted terms for this small vessel pathology, the rationale for making a clear distinction between CAA and sporadic non-amyloid microangiopathy is 2-fold. First, the two pathological processes are predominantly found in small vessels of different brain areas. In contrast to CAA, which typically affects the small arteries and arterioles in the cerebral cortex and overlying leptomeninges, sporadic non-amyloid microangiopathy predominantly affects the small perforating end arteries of the deep grey nuclei and deep white matter (Pantoni, 2010;Charidimou et al., 2016b). In parallel with this distribution of the underlying pathology, sporadic nonamyloid microangiopathy is commonly associated with cerebral microbleeds and spontaneous ICH in deep brain regions (e.g. basal ganglia, thalamus, and brainstem), whereas CAA is characterized by cerebral microbleeds and spontaneous ICH in a lobar distribution (cortical-subcortical), as well as cortical superficial siderosis. This allows for the clinical distinction of the two disorders based on the Boston diagnostic criteria (Knudsen et al., 2001;Smith and Greenberg, 2003;Linn et al., 2010;Martinez-Ramirez et al., 2015). The key neuropathological, clinical and imaging differences of CAA versus sporadic non-amyloid microangiopathy are summarized in Table 2. Of major clinical relevance, the two small vessel diseases have a different natural history and inherently distinct recurrent haemorrhagic risk in patients presenting with ICH: CAA-related lobar ICH carries a significantly higher risk for recurrence compared to deep ICH related to sporadic non-amyloid microangiopathy (Hill et al., 2000;Bailey et al., 2001;Viswanathan et al., 2006b;Weimar et al., 2011). This is particularly relevant for treatment decisions, especially antithrombotic use in the setting of small vessel disease (see 'Therapeutic aspects' section). Subdividing cerebral small vessel disease into CAA and non-CAA components may be an oversimplification as patients often have a mix of both pathologies. However, this distinction serves to distinguish the differences in the putative pathophysiology and different clinical and pathological consequences of the two diseases (Table 2).
Emerging concepts in cerebral amyloid angiopathy pathophysiology Perivascular drainage, vascular pathophysiology and cerebral amyloid angiopathy-related brain injury The pathogenesis of CAA and its downstream effects on the brain are incompletely understood and highly complex. However, there is little evidence that age-associated cerebrovascular accumulation in sporadic CAA is driven by amyloid-b overproduction. Instead, current evidence suggests that amyloid-b accumulation is driven largely by reduced peptide clearance (Deane et al., 2009;Mawuenyega et al., 2010). The small vessels of the brain appear to play a major role in this clearance process, both as sites of efflux across the blood-brain barrier (Zlokovic et al., 2010) and as the site of intramural perivascular drainage of amyloid-b from the brain interstitial fluid (Weller et al., 1998;Marin-Padilla and Knopman, 2011). Hence, the view of CAA as a protein elimination failure angiopathy has been devised (Carare et al., 2013;Carare and Kalaria, 2016) and provides the most conceptually compelling model to explain this failure and the pathophysiological basis of disease progression.
Details on the 'garbage truck of the brain', including brain pathways for lymphatic drainage and for the convective influx/glymphatic communication pathways are the focus of intensive scrutiny and not definitively established (Nedergaard, 2013;Tarasoff-Conway et al., 2015). However, the prevailing model of perivascular (peri-arterial) drainage of interstitial fluid relies on vessel pulsations as the motive force and biochemical interactions with basement membranes (Schley et al., 2006;Morris et al., 2016) that drive clearance countercurrent to blood flow (Boche et al., 2008;Carare et al., 2008). As perivascular drainage pathways progressively fail with age (possibly in combination with APOE e4 or e2 alleles or other pathological conditions), amyloid-b is increasingly trapped in the perivascular compartment allowing for increased aggregation and deposition along the basement membranes of small arteries. It is plausible that this failure in the cortex and resultant stagnation of interstitial fluid might lead to enlargement or dilation of perivascular spaces (also termed Virchow-Robin spaces) in the underlying white matter (Weller et al., 2015). In CAA, enlarged perivascular spaces often become visible on standard clinical MRI sequences, preferentially in the centrum semiovale (i.e. cerebral white matter) (Roher et al., 2003;Charidimou et al., 2013bMartinez-Ramirez et al., 2013;van Veluw et al., 2016). In line with this, a recent PET-MRI study suggested that high centrum semiovale perivascular spaces are associated with higher median cortical PiB retention across a wide range of cerebrovascular amyloid deposition (Charidimou et al., 2015a). A model on cause and effect for the perivascular drainage-amyloid accumulation hypothesis is not yet available. Nevertheless, perivascular drainage impairment potentially sets in motion a self-reinforcing pathway by which worsening vascular amyloid-b accumulation leads to activation of vascular injury pathways, impaired vascular physiology, and further increased parenchymal and vascular amyloidb accumulation (through a 'feed-forward loop' of reducing drainage efficiency) (Hawkes et al., 2011;Arbel-Ornath et al., 2013) (Fig. 2). Mouse model studies from multiple research groups have highlighted the potential importance of this pathway. Interruption of vascular pulsation by photocoagulation markedly reduces perivascular clearance (Arbel-Ornath et al., 2013) and increases local amyloid-b accumulation as plaques and CAA (Garcia-Alloza et al., 2011). The kinetics of CAA development are not well established, but vascular amyloid-b progresses mostly by propagation, which results in narrowing of unaffected gaps in small vessels (Robbins et al., 2006;Kimbrough et al., 2015) and expansion of amyloid-b from the basement membranes to progressively replace all tissue elements in the artery wall (Keable et al., 2016).
Another intriguing possibility is that the evolving pathophysiology of CAA is reflected and partly driven by specific alterations in the array of vascular gene expression (or vasculome) over time (Guo et al., 2012). Recent findings suggest that perturbations in the vasculome may mediate both brain function and injury. For example, under normal conditions, the endothelium can provide signals that support neurogenesis (Shen et al., 2004;Ohab et al., 2006), protect neurons against stress (Guo et al., 2008), and mediate homeostasis in oligodendrocyte precursor pools (Arai and Lo, 2009). When activated by injury, the endothelium can produce pro-inflammatory cytokines that magnify neuroinflammation and secondary injury (Pober and Sessa, 2007), disrupt endothelial tight junctions and basal lamina with blood-brain barrier breakdown (Rosenberg and Yang, 2007;Sandoval and Witt, 2008), and alter vessel physiology. It has also been suggested that amyloid might propagate and be transmitted in a prion-like fashion, but this is an active area of investigation (Jaunmuktane et al., 2015;Frontzek et al., 2016). At what 'tipping-point' CAA starts to propagate more CAA, potentially setting in motion this vicious cycle, is currently unknown.
The key molecular and physiological steps involved in damaging small vessels and retarding amyloid clearance, as well as the importance of these pathways in human cerebrovascular amyloid-b accumulation, have yet to be elucidated. At a neuropathological level, advanced CAA is associated with important morphological changes including loss of vascular smooth muscle cells (Dotti and De Strooper, 2009) and replacement with amyloid-b deposits, often markedly thickening the vessel wall (Mandybur, 1986;Vonsattel et al., 1991a;Greenberg et al., 2009a) or even vessel occlusion (Olichney et al., 1995) and loss of compliance leading to brittle, fragile vessels prone to microaneurysm formation and leakage (Attems et al., 2011). Cerebrovascular compliance is an almost entirely uncharacterized aspect of small vessel disease, but a candidate mediator of reduced perivascular clearance . The loss and distortion of the normal anatomical elements of the vessel wall in CAA presumably trigger the various types of structural focalmainly cortical-brain lesions such as ICH, cerebral microbleeds, siderosis and microinfarcts.

Capillaries and neurovascular dysfunction in cerebral amyloid angiopathy
Amyloid-b was recently shown to be highly toxic to pericytes, which, in turn, are important to blood-brain barrier function, basement membrane maintenance, and play a key role for neurovascular coupling (Hall et al., 2014). The distribution of capillary flows was also recently demonstrated to determine oxygen extraction efficiency (Jespersen and Ostergaard, 2012), implying that capillary pathology can be a source of tissue hypoxia and neuronal injury-even in the absence of flow-limiting pathology per se or structural alterations (Ostergaard et al., 2016). Whether these findings have specific implications in CAA has not yet been investigated directly, but data from the Alzheimer's disease field are compelling (Zlokovic, 2011;Nelson et al., 2016). However, they do raise the possibility that amyloid-b deposition at the level of the capillaries might lead to degenerative changes by two mechanisms: (i) interfering with various elimination pathways, including the perivascular drainage pathway, clearance at the bloodbrain barrier, or impairing the transendothelial clearance (Fig. 3); and (ii) affecting the metabolism of brain tissue (Kumar-Singh, 2009).

Cortical thinning in cerebral amyloid angiopathy
Of major interest, a recent study provided strong evidence of CAA-related cortical atrophy independent of Alzheimer's disease . Though the mechanism for this association is not well defined, the average cortical width of 63 non-demented CAA subjects versus 63 agematched healthy control subjects was reduced (2.17 AE 0.11 mm in CAA versus 2.31 AE 0.07 mm in controls, P 5 0.001)   (Fig. 3A). A similar finding was observed in subjects with Dutch-type hereditary CAA, a condition characterized by severe CAA but little or no Alzheimer's disease pathology (Natte et al., 2001). The average cortical thickness of 26 Dutch-type hereditary CAA subjects was significantly lower than age-matched healthy controls (2.31 AE 0.18 mm versus 2.42 AE 0.10 mm, P = 0.006) . Greatest thinning was noted in medial frontal and posterolateral brain regions, a somewhat different pattern from the primarily limbic and multimodal association regions with greatest thinning in Alzheimer's disease (Dickerson et al., 2009). The clinical implications of cortical thinning in CAA and its relation to cognitive impairment requires further study.

Altered vascular reactivity
CAA appears to have substantial effects on vascular function/physiology, which might result in more widespread brain injury (Smith and Greenberg, 2009;Pantoni, 2010). Amyloid deposition in the vessel wall may cause gliovascular unit impairment (Kimbrough et al., 2015), endothelial dysfunction (Grinberg et al., 2012), impaired autoregulation and blood-brain barrier disruption (Attems et al., 2011). One key candidate mechanism suggested by transgenic animal studies is impaired cerebrovascular compliance and reactivity to physiological stimuli (Niwa et al., 2000;Shin et al., 2007;Han et al., 2008;Park et al., 2013). The brain microvasculature forms interconnected vascular networks (a 2D network of pial arterioles that connects to a 3D network of subsurface microvessels) Shih et al., 2015); the practical consequence is that apparently unaffected microvessels might still contribute to the overall CAA-related tissue injury by indirect consequences from neighbouring amyloid-laden microvessels (Shih et al., 2013).
The best established physiological change in individuals with advanced CAA is reduced vasodilation to physiological stimuli. This finding has been demonstrated in studies measuring the functional MRI blood oxygen leveldependent (BOLD) response to visual stimulation (Dumas et al., 2012b;Peca et al., 2013) (Fig. 3B), which take advantage of the occipital predominance of CAA (Vinters and Gilbert, 1983). In non-demented CAA patients, the functional MRI response showed 27-28% reduced peak amplitude, 73% longer time to peak, and 42% longer time to return to baseline compared with elderly control subjects matched for age (Dumas et al., 2012b;Peca et al., 2013) (Fig. 3C). Altered BOLD response might be driven by vascular or neuronal dysfunction, but the findings that CAA patients and control individuals had similar visual evoked response amplitude (Peca et al., 2013) and that results were consistent in secondary analyses restricted to responding voxels only (Dumas et al., 2012a), suggest that the differences are largely driven by effects of CAA on the vessels themselves. A longitudinal analysis also showed declining amplitude in BOLD response to visual stimulation among CAA subjects (Switzer et al., 2016) (Fig. 3C). These studies also showed associations between impaired functional MRI response and white matter hyperintensity volume, suggesting mismatched blood supply versus tissue demand as a potential mechanism mediating CAA-related tissue damage. Another finding consistent with this possibility is that cortical thickness is negatively correlated with the time to peak BOLD response to visual stimulation (r = À0.4, P = 0.005 with adjustment for covariates) .

Microstructural damage and brain network alterations
CAA-related perfusion impairment may be responsible not only for subcortical white matter lesions (Gurol et al., 2013) but also for tissue microstructural changes detected by diffusion tensor imaging (Salat et al., 2006;Reijmer et al., 2015) and microinfarcts . Analysis (Fig. 4A) of the local network parameter node strength (the sum of fractional anisotropy-weighted connections attached to a given node) found locally altered network structure in CAA subjects compared to controls, particularly in occipital, posterior parietal, and posterior temporal cortex (Fig. 4B), resembling the distribution of CAA pathology . Network disturbances were associated with worse cognitive functioning and amyloid load on PET, providing direct links between vascular amyloid and impairments in white matter connectivity . Punctate diffusion-weighted imaging lesions (presumably corresponding to acute in cerebral amyloid angiopathy. (A) Regional differences in cortical thickness between patients with sporadic CAA and their age-matched controls (Fotiadis et al., 2016, reproduced with permission). Topographic surface maps were generated based on a general linear model (adjusting for age and sex) using a threshold of P 5 0.01 (with false discovery rate correction for multiple comparisons) . The resulting maps show the statistically significant regional differences in cortical thickness between the two groups. (B) Functional MRI visually activated region of interest of a representative probable CAA subject. The colour scale Figure 3 Continued denotes the Z-statistic for activation using the canonical haemodynamic response function. (C) Functional MRI measurements in response to visual stimulation in an elderly probable CAA patient (red error space) versus an age-matched control subject (blue error space). For the CAA patient, functional MRI was done at baseline (red error space) and again with the same scanner and protocol after one clinically asymptomatic year (yellow error space). The solid lines represent the change from baseline BOLD signal averaged over 16 cycles of visual stimulation (on 20 s, shaded region, then off 28 s), with standard deviations of the responses shown in blue, red and yellow spaces and the trapezoidal model fits, as previously described (Dumas et al., 2012a). In the CAA patient, the amplitude microinfarcts) in subcortical white matter have been demonstrated to produce focal changes in mean diffusivity and fractional anisotropy (Auriel et al., 2014), raising the idea that altered diffusivity may partly reflect the cumulative effect of cerebral microinfarcts or other focal lesions on cortical microstructure. Brain network alterations in patients with CAA (n = 33) appear to worsen measurably over just 1.3-year follow-up, progressing from posterior to frontal regions with increasing disease severity; this decline is associated with worse executive functioning (Beta = 0.41, P = 0.04) (Reijmer et al., 2016a). Of note, CAA patients included in all the aforementioned in vivo studies were symptomatic with high vascular lesion burden. A relevant question is whether advanced imaging techniques can also detect CAA-related abnormalities before the disease becomes symptomatic (Akoudad et al., 2013). This approach is currently limited, however, by our poor understanding of tissue-level pathological basis (Fig.  4C) of the various neuroimaging abnormalities discussed, making it difficult to link these large-scale measures to any specific pathogenic process or parenchymal injuries.

Towards mapping different cerebral amyloid angiopathy phenotypes
It is becoming increasingly evident that CAA is not a uniform, but rather a complex and very heterogeneous entity Figure 4 White matter injury in CAA detected with diffusion tensor imaging. (A) Whole brain fibre tract reconstruction from fractional anisotropy map reflecting the degree of anisotropic diffusion of water molecules. (B) Left hemisphere: visualization of a DTI-based brain network from a patient with severe CAA. Brain regions are depicted by nodes and fibre tracts between regions are depicted by edges. Right hemisphere: local differences in connectivity strength between patients with CAA compared to age-matched controls, as previously described . Results show that tracts projecting to occipital, parietal, and temporal regions are most affected, whereas tracts projecting to frontal and subcortical regions are relatively spared (darker nodes show greater differences from controls). (C) Regions marked in red were selected for histopathological evaluation of an autopsied CAA subject who underwent DTI scanning during life. The white matter near a region of reduced nodal strength showed considerable myelin loss on Luxol Õ Fast Blue stained sections relative to a region of relatively reserved nodal strength.
that can follow several different pathways (Figs 5 and 6). While sporadic CAA is commonly found in the elderly (Keage et al., 2009), it is usually mild and clinically silent. In the fraction of patients with CAA that will go on to develop clinical symptoms, the sentinel presentations of the disease vary (Greenberg et al., 1993;Maia et al., 2007), while each presentation is associated with its own cluster of biomarkers (Table 2). It follows that the clinical and imaging heterogeneity and phenotypes of the disease might be reflecting distinct neuropathological subtypes or patterns of cerebrovascular amyloid deposition (Fig. 5) and activation of diverging pathophysiological cascades. Mild CAA, a common finding in elderly and Alzheimer's disease brains, likely reflects the perivascular drainage impairment accompanying ageing. Severe CAA is generally assumed to result in the symptomatic sentinel presentations of the disease. However, the pattern on amyloid deposition in severe CAA varies greatly with differing anatomical patterns of deposition and degrees of accumulation and progression ( Fig. 6) (Keable et al., 2016). The spectrum ranges from amyloid co-localization with basement membranes, to replacement of smooth muscle cells by amyloid with some preservation of basement membrane elements, to complete replacement of the artery wall by amyloid (Fig. 6) (Keable et al., 2016). Yet, in some arterioles, amyloid deposition remains focal, even to the point of complete replacement of smooth muscle layers and basement membrane (Keable et al., 2016). The focal complete replacement of the vessel wall by amyloid might provide a potential site of weakness leading to haemorrhage. In other cases, deposition of amyloid is seen in the tunica adventitia surrounding the perimeter of the vessel (instead of the tunica media), which may represent an additional part of the lymphatic drainage pathway (Keable et al., 2016). Adding another dimension to the spectrum of the disease, these patterns of amyloid deposition may differentially affect leptomeningeal and cortical parenchymal vessels, as well as capillaries (Fig. 6). For example, a systems proteomic analysis demonstrated that clusterin (apolipoprotein J) and tissue inhibitor of metalloproteinases-3, co-localize with amyloid-b and seem to be upregulated in leptomeningeal (but not cortical) arteries from CAA patients compared to young and elderly controls (Manousopoulou et al., 2016).
The heterogeneity in these pathological patterns and clinical imaging presentations of CAA may be influenced by how different risk factors affect the process of perivascular clearance and brain injury during the disease process (Fig. 2). For example, APOE "4 alters the biochemical composition of basement membranes and is involved in various amyloid degradation systems, whereas APOE "2 promotes The two main trees (from right to left) depict Alzheimer's disease (AD) neurodegeneration, in which CAA is commonly found, and sporadic CAA. The third tree, in the background, represents sporadic non-amyloid microangioapathies. All three pathologies often co-exist in different combinations and severity in the ageing brain. Initially, different but overlapping pathophysiological pathways (roots of each tree in the grey box) can result in progressive vascular amyloid accumulation and pathophysiological alterations, which might culminate in structural vascular damage and brain injury. In extreme forms these might result in different neuroimaging and clinical phenotypes of the disease (different branches in each tree). At the one end we have the primarily haemorrhagic types of CAA, predominantly characterized by either multiple lobar cerebral microbleeds (CMB) (i.e. 'microbleeders') or cortical superficial siderosis (cSS) ('superficial bleeders'), and/or symptomatic ICH (i.e. 'macrobleeders'); at the other end we have the less haemorrhagic phenotypes of CAA, clinically expressed with more cognitive impairment and probably more strongly associated with cortical microinfarcts (CMI), with or without Alzheimer's disease. Many patients can have intermediate phenotypes of the disease. The APOE " genotype and vascular risk factors likely influence all the critical steps and different directions taken during the course of the disease in the ageing brain. Ab = amyloid-b; TFNEs = transient focal neurological episodes; WMH = white matter hyperintensity. structural vasculopathic changes in amyloid-laden vessels, making them more prone to rupture. Mid-life hypertension might alter the biophysical forces acting upon the arterial wall, modifying the motive force for perivascular clearance but also the structural integrity of the vessels (in combination with sporadic non-amyloid microangiopathic alterations). Other potential risk factors and comorbid conditions that result in vessel rupture associated with CAA remain unclear. Of note, the degree of vessel wall replacement by amyloid required before the vessel ruptures is currently unknown.
A number of observations from clinical-MRI CAA cohorts can shed further light on the concept of different disease phenotypes. An interesting study suggested that multiple lobar microbleeds associated with CAA-related pathology might in some respects differ from macrobleeds, since they have a bimodal (rather than continuous) size distribution on MRI (Greenberg et al., 2009a). In a patient subset that underwent autopsy, those with high microbleed counts ('microbleeders') had increased wall thickness of amyloid-laden vessels compared to patients with relatively low microbleed counts and lobar ICH ('macrobleeders') Figure 6 Brain anatomical patterns, evolution and neuropathological phenotypes of cerebrovascular amyloid-b deposition in advanced CAA. (A) Schematic representation of the superficial cortical small vessel disease system in the brain, typically affected by CAA. Leptomeningeal arterioles give off short (S) penetrating arterioles ('cortical') reaching three different depths in the cortex (i.e. cortical layer III, V and the grey-white matter junction, S1-3, respectively), while long penetrators ('medullary', not usually affected by CAA) continue into the subcortical white matter. In advanced disease, often decreasing severity of CAA is seen between leptomeningeal, superficial and deeper layers in the cortex. (B) Regional severity of CAA with more heavy involvement of occipital and temporal lobes. Severe CAA probably progresses in a posterior-to-anterior fashion in the brain. (C) Neuropathological patterns spectrum of severe amyloid-b deposition. C(a)-C(d). Amyloid colocalization with basement membranes, to replacement of smooth muscle cells with some preservation of basement membrane elements, to complete replacement of the artery wall. C(e) More pronounced CAA-related vessel wall damage resulting in vasculopathic changes. C(f) Very focal amyloid deposition, but with complete replacement of smooth muscle and basement membrane layers at the point of vessel wall cracking, potentially providing a site of weakness and haemorrhage. C(g) Amyloid-b deposition in the tunica adventitia surrounding the perimeter of the vessel (instead of the tunica media). C(h) Capillary CAA, which might range from mild to severe, with dysfunction of pericytes and endothelial cells to complete capillary occlusion. A-C could potentially influence the spectrum of clinical imaging phenotypes of CAA, as neuropathological patterns of amyloid deposition can differentially affect leptomeningeal and cortical parenchymal vessels, with distinct anatomical patterns of deposition, degrees of accumulation, progression and pathophysiological consequences during the disease course. [C was redesigned and modified based on data and figures from (Keable et al., 2016), doi: 10.1007/s00401-016-1555-z, under Creative Commons Attribution License (CC BY)]. (Greenberg et al., 2009a). A recent MRI-neuropathological studies recruited 105 patients with CAA pathologic confirmation and in vivo MRI and compared pathologic, imaging, and APOE genotype data between subjects with versus without symptomatic ICH. The two groups had similar CAA burden, almost all pathology samples had neuritic plaques, whereas neurofibrillary tangles were more commonly present in the patients without intracerebral bleeds (87% versus 42%, P 5 0.0001). Disseminated cortical superficial siderosis was considerably more common in patients with ICH (33.3% versus 5.9%, P 5 0.0001) and was associated with APOE "2 (but not APOE "4) (Charidimou et al., 2015c). Surprisingly, cortical superficial siderosis, which confers a high risk of future lobar ICH, is not associated with lobar microbleeds (Charidimou et al., 2013a;Shoamanesh et al., 2014). A plausible explanation is that siderosis predominantly reflects a severe leptomeningeal pattern of amyloid deposition rather than deeper cortical CAA, which manifests itself more with microbleeds. Although studies of CAA phenotypes often focus on the extreme forms (such as microbleeders, macrobleeders, siderobleeders, etc.), many patients likely fall into a mixed category in which various CAArelated vascular processes co-exist.
Although literature to support firm conclusions remains extremely limited, a 'phenotype approach' in CAA may provide a conceptual model allowing greater reliance on different clusters of clinical and pathophysiologically relevant MRI biomarkers in future CAA trials and patient management (Fig. 6) (Greenberg et al., 2014).

Implications for patient management in cerebral amyloid angiopathy
The management of acute CAA-related ICH is the same as any spontaneous ICH, and it is summarized in the American Heart Association/American Stroke Association (Hemphill et al., 2015) and the European Stroke Organization (Steiner et al., 2014) guidelines. To date, there is no disease-specific CAA treatment, and hence the long-term management of CAA patients is centred on the prevention of incident or recurrent haemorrhagic events and dementia. In this context, an important first step in decision-making for certain drug use or avoidance (antithrombotics, statins etc.) and other interventions is identifying the probability that a given patient with spontaneous ICH has CAA. In this setting, an accurate assessment of haemorrhage risk based on the presumed cause of ICH and/or the predominant type/severity of the underlying haemorrhage-prone microangiopathy. A key difference between the two broad types of small vessel disease and aetiologies of spontaneous ICH is the much higher annual recurrence rate in CAA-related lobar ICH compared to sporadic non-amyloid microangiopathy-related ICH ($10%/year versus 2-3%/year, respectively). Furthermore, the risk of incident dementia after ICH appears to be more than two times higher in patients with lobar (incidence at 1 year: 23.4%; 95% CI: 14.6-33.3) than for patients with non-lobar ICH (incidence at 1 year: 9.2%; 95% CI: 5.1-14.7) (Moulin et al., 2016).
A common approach for CAA patients with a previous history of lobar ICH is to avoid antithrombotic treatment as much as possible, as they might increase the risk of recurrence (Eckman et al., 2003;Hofmeijer et al., 2015). Even for strong indications such as non-valvular atrial fibrillation, oral anticoagulation with warfarin may be generally unsafe following CAA-related ICH (Eckman et al., 2003), although thus far, large observational data are missing. American Heart Association/American Stroke Association guidelines recommend avoiding oral anticoagulation in patients with lobar ICH, but to consider resuming oral anticoagulation in patients with non-lobar ICH (Hemphill et al., 2015). The role of antiplatelet agents and their systematic discontinuation or not in CAA-related ICH patients remains a hotly debated topic (Al-Shahi Salman and Dennis, 2014;Falcone and Rosand, 2014). The RESTART randomized trial (http://www.restarttrial. org/) is underway to provide some answers on whether to avoid antithrombotic use after an ICH. Another option for those CAA-related ICH patients requiring anticoagulation might be newer oral anticoagulants, which have been shown to have lower overall ICH risk. For example, apixaban has a similar intracranial bleeding risk profile as aspirin, but is more effective at reducing ischaemic stroke risk (Connolly et al., 2011). The relative risks are less clear for individuals with asymptomatic lobar microbleeds or cortical superficial siderosis only, fulfilling the Boston criteria for CAA. In this situation, clinicians should carefully weigh risk-benefit for recurrent ischaemic stroke versus haemorrhagic complications. Larger, more definitive studies (Charidimou et al., 2015e) and randomized trials are required to settle these important questions.
Although CAA is not thought to be directly linked to hypertension, strict control of blood pressure within the normal range is recommended, especially after an ICH. A subgroup analysis of the PROGRESS trial (Arima et al., 2010) demonstrated that lowering blood pressure with the antihypertensive drug perindopril (with or without indapamide) reduced the risk of probable CAA-related ICH recurrence by 77% (95% CI, 19%-93%) over a follow-up period of 3.9 years. Recently, inadequate blood pressure control in ICH patients from a large single centre observational study was associated with hazard ratios of 3.53 (95% CI, 1.65-7.54) for recurrent lobar ICH and of 4.23 (95% CI, 1.02-17.52) for non-lobar ICH when compared to those with adequately controlled blood pressure (Biffi et al., 2015). These and other observations strongly support the need for strict and continued blood pressure control in CAA patients, which might impact outcomes such as recurrent ICH, cognitive impairment, or progression of small vessel disease pathology. Finally, supportive care, lifestyle modifications and, in patients with recent CAA-related ICH, cognitive rehabilitation may maximize function and enhance recovery.

Future perspectives
There is currently a growing interest in CAA with an explosion of publications in the last 15 years and dense collaboration networks between research teams in the field (Charidimou et al., 2016a). International efforts aimed at translating mechanistic insights into novel treatment approaches (Al-Shawi et al., 2016), including an ongoing first-of-kind CAA anti-amyloid immunotherapy trial with ponezumab, an anti-amyloid-b 40 selective antibody (NCT 01821118) are underway (Greenberg et al., 2014). The rationale for anti-amyloid immunotherapy in CAA is to attenuate amyloid accrual in cerebral vessels and by doing that restore vascular reactivity in the short-term and prevent the occurrence of clinical symptomatology in the longterm (Bales et al., 2016). Rapidly accumulating data on in vivo detection of the pathogenic steps involved in CAA offers substantial promise for future trials aimed at identifying disease-modifying treatments (Saito and Ihara, 2014) for this largely untreatable disease. Nonetheless, there are many remaining 'known unknowns' on all the postulated steps in CAA pathogenesis and neurological dysfunction, ranging from vascular amyloid deposition itself to physiological alterations, haemorrhagic and non-haemorrhagic (cortical and subcortical) structural damage, symptomatic ICH, and presentations without major haemorrhage, including cognitive dysfunction. It is anticipated, that shedding light on these aspects may also have implications for other small-vessel disease processes in the ageing brain and vascular cognitive impairment.