Clinical, imaging, and pathological heterogeneity of the Alzheimer's disease syndrome

With increasing knowledge of clinical in vivo biomarkers and the pathological intricacies of Alzheimer's disease (AD), nosology is evolving. Harmonized consensus criteria that emphasize prototypic illness continue to develop to achieve diagnostic clarity for treatment decisions and clinical trials. However, it is clear that AD is clinically heterogeneous in presentation and progression, demonstrating variable topographic distributions of atrophy and hypometabolism/hypoperfusion. AD furthermore often keeps company with other conditions that may further nuance clinical expression, such as synucleinopathy exacerbating executive and visuospatial dysfunction and vascular pathologies (particularly small vessel disease that is increasingly ubiquitous with human aging) accentuating frontal-dysexecutive symptomatology. That some of these atypical clinical patterns recur may imply the existence of distinct AD variants. For example, focal temporal lobe dysfunction is associated with a pure amnestic syndrome, very slow decline, with atrophy and neurofibrillary tangles limited largely to the medial temporal region including the entorhinal cortex. Left parietal atrophy and/or hypometabolism/hypoperfusion are associated with language symptoms, younger age of onset, and faster rate of decline - a potential 'language variant' of AD. Conversely, the same pattern but predominantly affecting the right parietal lobe is associated with a similar syndrome but with visuospatial symptoms replacing impaired language function. Finally, the extremely rare frontal variant is associated with executive dysfunction out of keeping with degree of memory decline and may have prominent behavioural symptoms. Genotypic differences may underlie some of these subtypes; for example, absence of apolipoprotein E e4 is often associated with atypicality in younger onset AD. Understanding the mechanisms behind this variability merits further investigation, informed by recent advances in imaging techniques, biomarker assays, and quantitative pathological methods, in conjunction with standardized clinical, functional, neuropsychological and neurobehavioral evaluations. Such an understanding is needed to facilitate 'personalized AD medicine', and eventually allow for clinical trials targeting specific AD subtypes. Although the focus legitimately remains on prototypic illness, continuing efforts to develop disease-modifying therapies should not exclude the rarer AD subtypes and common comorbid presentations, as is currently often the case. Only by treating them as well can we address the full burden of this devastating dementia syndrome.

current discussion. Instead this review will focus on sporadic cognitive subsyndromes.
Th e growing recognition in recent years of AD variability refl ects increasing understanding of its complexity, building upon convergent streams of evidence from clinical, imaging, and pathological studies. While it is legitimate to focus on prototypic illness, as this represents the majority of cases and thus the bulk of disease burden, understanding the rarer AD subtypes and the infl uence of common comorbidities may open new targets for therapy and better allow the tailoring of existing ones -personalized medicine. Concurrently, know ledge of AD's syndromic inhomogeneity may disentangle its confounding eff ects in clinical trials, as eff orts to exclude atypical subtypes are not always successful.

Origins of the AD heterogeneity concept: a historical perspective
Th e observation that AD demonstrates phenotypic hetero geneity is not new. In 1969, McDonald [10] identifi ed two distinct subgroups among dementia patients in a chronic geriatric hospital setting. Using simple tests of memory, parietal function, and aphasia, he noted that some had diffi culties predominantly with praxis, visual construction, and cortical sensation. Th ey exhibited more severe progression on follow-up. Th ese he termed the 'parietal group' . Other patients had predominantly memory dysfunction, later age of onset, and slower disease progres sion. Th ese he termed 'benign memory dysfunction of aging' . Unfortunately, this early demonstration of AD heterogeneity failed to generate widespread recognition for subtype variability. In large part this was due to the prevailing theory of the time, which held that clinical variation arose from observing the disease at diff erent stages of progression (phase hypothesis), rather than truly distinct disease phenotypes (subtype hypothesis) [11].
Additional support for the subtype hypothesis would later be provided by early positron emission tomography (PET) studies. Th ese demonstrated that just as there were clinically distinct profi les of AD, there were distinct topographic patterns of brain hypometabolism. Asymmetry in PET imaging among AD subjects [12,13] was consistently associated with greater language impairment if the left hemisphere was more aff ected. Conversely, visuospatial impairment predominated in those with mainly right parietal hypometabolism [14,15]. Th is variability exhibited good anatomic correlation. Left angular hypometabolism, for example, was associated with Gerstmann's syndrome in AD [14]. Importantly, longitudinal follow-up demonstrated that these diff erent syndromes remained distinct over time [16], and with clinical worsening of disease [17].
Th e one remaining observation that still supported the phase hypothesis was that these variant cases were associated with an earlier age of onset [18,19]. Th is was later addressed in studies taking advantage of standardized neuropsychological assessments. Using the CERAD (Consortium to Establish a Registry for Alzheimer's Disease) database, Fisher and colleagues [20] demonstrated neuropsychologically defi ned subtypes of AD. Subjects with predominantly anomia, impairment of construc tional praxis, and mixed impairments mirrored the left, right, and general subgroups of the earlier PET studies, respectively. Importantly, Fisher's subtypes had the same age of onset, strongly suggesting that these observed diff erences were not merely alternative stages, but true variants of AD. As with PET studies, these subtypes remained distinct over longitudinal follow-up, again supporting the idea of true distinct subtypes of disease [21].

Typical Alzheimer's disease
Prototypic AD is a late-onset AD syndrome with amnestic impairment predominating in association with hippocampal and temporal-parietal atrophy and/or decreased perfusion/metabolism [22]. It is the most commonly observed AD phenotype in the clinical setting and serves as a good starting point against which the rare AD subtypes can be compared. Clinically, memory decline is accompanied by similar worsening in other cognitive domains, setting typical AD apart from temporal variant AD where memory decline occurs in isolation, and the other variants wherein non-amnestic presentations predominate. Relatively symmetric and generalized atrophy and hypometabolism/hypoperfusion distinguish typical AD from the more focal topography of temporal variant (hippocampal), frontal variant (frontal), language variant (left parietal), and visuoperceptive variant (right parietal) AD. Typical AD progresses more slowly than language, visuoperceptive, and frontal variant AD. Conversely, deteri ora tion is quicker than temporal variant AD, perhaps owing to greater cognitive reserve in these individuals (consistent with their older ages of onset), the more focal mesiotemporal distribution of pathology that tends to spare cortical areas, or some combination of both.

Temporal (pure amnestic) variant Alzheimer's disease
Focal temporal lobe dysfunction, pure amnestic AD, and temporal variant AD all refer to the late-onset AD syndrome of isolated episodic memory impairment with notably slow decline ( Figure 1) [23]. Single-photon emission computed tomography (SPECT) imaging in temporal variant AD demonstrates hypoperfusion limited to the mesiotemporal lobes, while the temporal-parietal changes seen in typical AD are absent [24]. Longitudinal studies of temporal variant AD individuals demonstrate slow or no change in Mini-Mental State Examination scores, and even when memory is signifi cantly impaired, visuospatial and executive function remain borderline to normal [23,25]. Pathologically, studies have demonstrated a subgroup of patients with plaques and neurofi brillary tangles limited to the limbic regions with little or no spread to the neocortical areas [26]. Clinically these individuals have a later age of onset and slower rates of cognitive decline. Although the genetic factors contri buting to temporal variant AD remain unknown, there is some evidence to suggest that the apolipoprotein E (APOE) ε4 allele is absent [23]. Unique among atypical AD variants, temporal variant AD is a late-onset AD syndrome, and may present even later than typical AD.

Left (language) variant and logopenic progressive aphasia
Language variant AD is often an early-onset AD (EOAD) syndrome of gradually worsening non-fl uent speech typifi ed by signifi cant agrammatism, phonemic para phasias, relative preservation of memory, and often atrophy of the left perisylvian region on imaging ( Figure 2) [27][28][29]. Th ese individuals have pathologically confi rmed AD with a topographically atypical distribution of neurofi brillary tangles predominantly within the left neocortex, sparing in some cases the hippocampus.
Th e early non-fl uent language impairment of this subtype distinguishes it from the language syndrome of later stage typical AD, which is generally fl uent in nature, with anomia, semantic paraphasias, progressing to surface dyslexia and jargon speech [30,31].
Th is non-fl uency also distinguishes language variant AD from the second atypical AD language syndrome, logo penic progressive aphasia (LPA). In LPA, speech rate is slowed but grammar and articulation are preserved [32]; rather, impaired repetition typifi es LPA [33]. LPA is commonly associated with AD pathology [34], demonstrates left posterior temporal and inferior parietal hypoperfusion [32], and strong association with β-amyloid deposition on Pittsburgh compound B PET [35].
Despite their shared features, both being language predominant syndromes, the basis for these two forms of diff erent aphasic manifestations is unclear.

Right (visuoperceptive) variant (visuoperceptive AD)
Visuospatial dysfunction does not commonly occur as the initial or predominant symptom in AD. When it does, it may portend the onset of PCA ( Figure 3) [36]. In other cases, visuospatial dysfunction suggests a subtler, nonmemory variant, associated with greater right versus left hemisphere pathology and atrophy ( Figure 4), a distinct pattern that is maintained over time and with disease progression [14,15,17,21].
As with language variant AD and LPA, the relationship between visuoperceptive AD and PCA remains relatively unknown. Also unknown is whether individuals with visuoperceptive impairment represent a prodromal stage of diff use Lewy body disease, rather than a true AD subtype, as sometimes happens anecdotally. While it is true that Lewy bodies can co-occur in PCA alongside AD pathology with clinically evident hallucinations and parkinsonism, AD changes can also be seen in isolation, with topography being the only apparent driver of phenotype [36]. Pathologic and biomarker imaging studies may help resolve some of these questions.

Frontal (executive) variant (frontal variant AD)
Frontal variant AD is an extremely rare EOAD subtype, associated with signifi cant frontal cognitive and behavioural symptoms ( Figure 5), fi rst described by Johnson and colleagues [37], who found 3 cases among 63 individuals with pathologically confi rmed AD. Alladi and colleagues [34] likewise identifi ed only 2 instances among their 100 case series, of whom only one had a true dysexecutive syndrome, the other having only behavioural features. Pathologically there is a predominance of NFTs in the frontal regions (ten-fold increase over typical AD), with comparable loading in the entorhinal cortex and other regions [37]. Amyloid plaques and the lack of cell loss, microvacuolarization, and gliosis in layers II and III distinguish frontal variant AD from FTD. Biochemically, frontal variant AD shows focally reduced calcium-independent phospholipase A2 activity within the frontal regions compared to typical AD on post-mortem protein assay (8.8 pmol/mg versus 15.2 pmol/mg), a neuronalspecifi c isoform of phospholipase and indirect marker of neuronal health [38]. In one case report, cerebrospinal fl uid (CSF) amyloid beta (Aβ) was found to be decreased, although tau levels were only borderline [39]. In part because of its rarity, the clinical identifi cation of frontal variant AD has proven troublesome. One study used the Frontal Behavioural Inventory to defi ne a 'high frontality' AD group based on the presence of frontal symptoms such as apathy, aspontaneity, loss of empathy (negative symptoms) and disinhibition, utilization behaviour, and alien limb phenomenon (positive symptoms) [40]. However, except for three items on the Frontal Behavioural Inventory -hyperorality, perseveration, and aspontaneity -there was no distinction between Woodward's 'high frontality' group and FTD and there were no between-group diff erences on global cognition screening tests. Furthermore, Woodward and colleagues in another study [41] found that frontal variant AD can be associated with co-occurring FTD pathology. In light of this overlapping pathology and diagnostic uncertainty, CSF biomarker profi les and amyloid imaging will likely be needed to convincingly demonstrate frontal variant AD in a cohort large enough to perform the necessary neuropsychological, imaging, and genetic studies to defi ne this subtype.

Age of onset
EOAD accounts for 32% of atypical, that is, nonamnestic, cases of AD, in contrast to only 6% of typical AD [42]. In retrospect, this is consistent with earlier studies, wherein atypical subgroups of AD were consis tently associated with younger age. As alluded to previously, these early-onset individuals evince a more aggressive disease course, in distinction from the more gradual progression of typical AD, and in contrast to the very slow decline of temporal variant AD. Th e mechanism and signifi cance of this association remain unclear.

Disease topography
Th e distribution of AD pathology and associated atrophy varies among individuals [43] and may aff ect phenotype. For example, histopathological asymmetry with associated left (language) and right (visuospatial) syn dromes similar to those found in PET studies have been described [44].
Conversely, focal cortical, non-amnestic presentations can be associated with underlying AD in 34% of cases, including bilateral PCA, aphasia, behavioural executive syndromes, and cortical basal syndrome, all of which tended to be EOAD [34].
In one recent study, Armstrong and colleagues [26] used principal components analysis to identify three clusters of NFT and senile plaque distribution: cortical (cingulate gyrus, gyrus rectus, orbital frontal gyrus, occipital lobe), deep grey (thalamus, nucleus basalis of Meynert, striatum), and limbic (ventral tegmentum, raphe, amygdala). Similarly, Murray and colleagues [45] described three pathologically distinct patterns of AD in a large cohort of 889 cases: hippocampal-sparing, limbicpredominant, and 'typical' . Th ese subdivisions were based on NFT and senile plaque burden, as defi ned by thiofl avin-S counts, comparing between mesiotemporal lobe (CA1, subiculum) and cortical (middle frontal, inferior parietal, superior temporal association areas) regions of interest. Hippocampal-sparing AD occurred in younger individuals (mean age 72 years, versus 79 for typical and 86 for limbic-predominant AD), consistent with the asso ciation between atypicality and EOAD. Hippocampal-sparing AD revealed the fastest rate of cognitive decline (-4.8 on Mini-Mental State Examination per year, versus -2.8 for typical and -1.4 for limbicpredominant AD). In keeping with their neocortically predominant patho logi cal burden, hippocampal-sparing cases were more likely to have an atypical, non-amnestic clinical onset (30% of cases), versus individuals with typical pathological distribution, where this occurred less frequently (17%).

Genetics
ApoE genotype may be one important factor contributing to heterogeneity in sporadic AD, as non-ε4 status among EOAD patients correlates with atypicality [46]. What underlies this relationship is unknown; perhaps the absence of ε4, rather than the presence of ε2 or ε3, is important. ApoE ε4 is associated with greater hippocampal atrophy [47], suggesting that symptoms in noncarriers may instead refl ect damage to areas normally eclipsed by hippocampal pathology (for example, parietal, temporal or frontal regions). Th is is in keeping with the 'cortically predominant' topographies of disease observed among non-ε4 EOAD carriers. Th is greater degree of cortical damage could be reasonably expected to result in signifi cant, wide-spread neurologic dysfunction, potentially explaining the observation that such individuals experience a more rapid clinical decline. Alternatively, this same 'hippocampal eff ect' of ApoE ε4 may mask the infl uence of other genetic and epigenetic factors. If so, such factors may have a greater role in the absence of ε4; their variability in turn explaining the greater heterogeneity of non-ε4 EOAD.
Another consideration is the role of autosomal dominant mutations causing familial AD. Although a complete discussion of genotype-phenotype correlations in familial AD is beyond the scope of this review, mutations in PSEN1 and amyloid precursor protein (APP) can both produce non-amnestic, atypical EOAD [48,49]. For example, PSEN1 can result in non-fl uent aphasia [50] in addition to more typical amnestic AD, while APP can be associated with severe cerebral amyloid angiopathy present ing with hemorrhage and seizures along with memory decline [51]. Th e same mutation may even result in diff erent syndromes. PSEN1 has been described in spastic paraparesis, frontotemporal dementia, myoclonus with seizures, and predominantly psychiatric presen tations [52]. Even within the same family, the APP mutation presented with bradykinesia and hallucinations in one individual, memory and behavioural changes in another, and memory decline followed by intracerebral hemorrhage from angiopathy in a third [53].
Ultimately, imaging-genetic endophenotype studies may provide a link between genetics and disease topography by elucidating those areas of the brain most associated with known and potential pathological genotypes. Should these patterns correlate with the topography of syndromic phenotypes of AD, it may lend support that genotype underlies at least some of these phenotypic variations. Until such links can be established, and given the variability with which genetic mutations/ polymor phisms can present, decisions around genetic testing when a suspected case of atypical AD is encountered must be adjudicated on a case by case basis.

Co-occurring pathology: Lewy body pathology
Lewy bodies composed in part of alpha-synuclein aggregates are the hallmark of Lewy body disease, presenting in prototypic cases with visual hallucinations, extra pyramidal symptoms, and marked clinical fl uctuations [54]. Th ese same alpha synuclein inclusions can be found in pathologically confi rmed AD, with neocortical Lewy bodies occurring in 10 to 30% of cases at autopsy [55,56]. Th ese co-occurring Lewy bodies also appear associated with AD proven cases of PCA [29,57]. More generally, the presence of Lewy bodies in AD is associated with more perceptual impairment, though milder than what occurs in 'pure' diff use Lewy body disease [58].
As the degree of WMH varies widely, investigators have explored whether a dose-eff ect relationship for WMH exists, and if so what minimum degree of WMH is necessary for cognitive changes to be observed. To this end, Boone and colleagues [80] identifi ed a 'threshold' of >10 cm  for cognitive changes to be observed. DeCarli and colleagues [81] demonstrated similar fi ndings using PET, showing that a threshold of >0.5% of brain parenchymal fraction was signifi cantly associated with poorer performance on executive function and mental fl exibility tasks (phonemic fl uency, Trails B). Th ey also demonstrated an association with worsened visual memory, suggesting that the infl uence of WMHs is potentially more widespread. Importantly, even when only accounting for those cases where WMH met or exceeded the minimum 'threshold' for associated clinical symptoms, co-pathology was common (6 to 20%) [60,80,81], underscoring its import ance in overall AD phenotype.
A major challenge to investigating independent contributions of WMHs in AD is disentangling their contributory role to cerebral atrophy. Controlling for atrophy can eliminate the observed infl uence of WMHs on cognition [81,82], although Swartz and colleagues [83], using factor analysis in a dementia population, discerned an independent association between WMH burden on executive function and memory even after atrophy was accounted for. Furthermore, vasculopathy, especially lacunar infarcts, appears to contribute to the expression of AD dementia in pathology series [84,85].

Co-occurring pathology: other pathologies
TDP-43 (TAR DNA-binding protein 43) inclusions and agyrophillic grains are two other pathologies that have been associated with AD pathology and that may nuance . One year prior, axial single-photon emission computed tomography (SPECT) at the corresponding level (bottom right) demonstrated mild parietal hypoperfusion bilaterally (red arrows). The patient died 9 years later and autopsy confi rmed Alzheimer's disease (Braak V/VI), with diff use atherosclerosis throughout the white matter and basal ganglia, lacunar infarcts, remote microhaemorrhages and cortical microinfarcts. There had been no history of visual hallucinations, but parkinsonism developed very late into the disease course, and he was also found to have diff use Lewy bodies within the brain stem and cingulum.
the phenotype of AD [86]. A full review of their association is beyond the scope of this review, although given their association with FTD, one might expect that they would result in a blending of FTD symptomatology with prototypic AD.

Future directions
Framing heterogeneity and complexity in the AD syndrome AD phenotype is strongly aff ected by age of onset, genetic profi le, and comorbidities. AD subtypes refl ect the interactions between these three axes. When the onset is later, ApoE ε4 is present, and comorbidities absent, the resulting pattern is generally prototypic AD. Changes along any of these three axes shift the observed phenotype towards one of the rarer AD variants (Figure 7).
Several important questions remain. Frontal variant AD appears to be the rarest variant and the potential role of WMHs in its expression needs to be further explored, in particular the role of venous collagenosis, which may contribute to executive control network dysfunction. Th e relationships between visuoperceptive AD and PCA, and between language variant AD and LPA remain unclear. ApoE ε4 is hypothesized to be absent in the early onset subtypes: frontal, visuoperceptive, and language variant AD, but its relative absence in temporal variant AD, the only late onset subtype, is also worthy of further investigation. Finally, the role of rarer co-pathologies such as TDP43, and genetic polymorphisms, other than ApoE, and epigenetic factors on clinical profi le remain to be explored.

Translational implications: diagnosis and therapy
Phenotypic heterogeneity among AD subtypes and copathology may have particular importance for biomarkerbased ante mortem diagnosis. Factors such as ApoE status, gender, age, education, and brain size appear associated with diff erences in CSF Aβ 42 levels [87], with education inversely correlated with CSF Aβ 42 in early disease [88]. Such fi ndings suggest that, at least in the case of CSF Aβ 42 , biomarkers must be interpreted in the context of phenotype. Th is interpretation in turn requires a greater understanding of AD phenotypic variation than is presently available. As atypical subtypes are relatively rare, multi-site studies specifi cally addressing them are urgently needed to delineate the full spectrum of AD and its interaction with biomarkers.
Beyond the clinical need for diagnostic certainty, the ability to recognize and control for phenotypic variation is important to continuing clinical trials design. Th e very slow rate of decline in temporal variant AD is but one example of how subtype heterogeneity can confound results. More generally, as it appears that all AD subtypes exhibit divergent trajectories of symptom progression from prototypic disease, characterizing the onset and decline across the AD syndrome spectrum [89,90], including comorbid cases, is a priority. Similarly, it is unknown how often an initially typical AD case evolves into an atypical subtype over time, or vice versa. As the occasional inclusion of an atypical individual in clinical trials may thus be unavoidable, understanding these subtypes may allow researchers to disentangle their infl uence from their fi ndings.
Phenotypic variation may likewise aff ect response to therapeutic strategies, requiring a personalized medicine approach that will only be possible with a greater understanding of AD subtypes and their causes. For example, the presence of comorbid WMHs involving peri-insular cholinergic pathways in a case-controlled cohort study was associated with more favourable response to cholinesterase therapy [91], in contrast to more generally distributed WMHs, which show no such association. Cholinergic therapies may therefore be particularly appropriate in such co-morbid cases. Another example is the observation that synapse loss, a key correlate of cognition, is greater in early onset AD [92,93], implying a more aggressive neuronal degeneration. Clinical trials in these subgroups would therefore have to target symptoms much earlier than would normally be done for prototypic disease, in order to address the more severe cellular damage of EOAD. Along similar lines, it has been observed that early onset AD is associated with more widespread neurotransmitter dysfunc tion [94], aff ecting noradrenaline, γ-amino-butyric acid, and somatostatin levels in addition to acetylcholine. Hence, combination therapy that restores multiple neurotransmitters may be more symptomatically eff ective than cholinergic agents alone in EOAD. While the medications for such a clinical trial are already available, the lack of clear diagnostic criteria for EOAD subtypes hampers implementation, further underscoring the urgent need for further research in AD heterogeneity.

Conclusion
Prototypic and atypical AD subtypes exist along continuums of age, genotype, and co-pathology within the AD syndrome, presenting challenges and opportu nities for both researchers and clinicians. While the pursuit of treatments and salient criteria for the more common AD prototype understandably remains a priority, these rarer subtypes pose a substantial burden of disease faced by those aff ected by them and their caregivers. Furthermore, unless atypical variants are understood and recognized, controlling for their potentially confounding eff ects in clinical trials will be more diffi cult, hindering treatment development even for prototypic disease.