Recent Advances in Imaging of Preclinical, Sporadic, and Autosomal Dominant Alzheimer’s Disease

Abstract

Observing Alzheimer’s disease (AD) pathological changes in vivo with neuroimaging provides invaluable opportunities to understand and predict the course of disease. Neuroimaging AD biomarkers also allow for real-time tracking of disease-modifying treatment in clinical trials. With recent neuroimaging advances, along with the burgeoning availability of longitudinal neuroimaging data and big-data harmonization approaches, a more comprehensive evaluation of the disease has shed light on the topographical staging and temporal sequencing of the disease. Multimodal imaging approaches have also promoted the development of data-driven models of AD-associated pathological propagation of tau proteinopathies. Studies of autosomal dominant, early sporadic, and late sporadic courses of the disease have shed unique insights into the AD pathological cascade, particularly with regard to genetic vulnerabilities and the identification of potential drug targets. Further, neuroimaging markers of b-amyloid, tau, and neurodegeneration have provided a powerful tool for validation of novel fluid cerebrospinal and plasma markers. This review highlights some of the latest advances in the field of human neuroimaging in AD across these topics, particularly with respect to positron emission tomography and structural and functional magnetic resonance imaging.

Supplementary Information

The online version contains supplementary material available at 10.1007/s13311-021-01026-5.

Introduction

Alzheimer’s disease (AD) is the most common form of dementia, expressing a progressive neurodegenerative condition that is primarily defined by the accumulation of mis-aggregated β-amyloid (Aβ) peptides in extracellular space and hyperphosphorylation of tau protein to form neurofibrillary tangles (NFT), which results in neurodegeneration and objective cognitive decline[1, 2]. This review will cover some of the most recent advances in human neuroimaging of Alzheimer’s disease, with a focus on positron emission tomography (PET) markers of Aβ and tau, structural atrophy, and functional dysconnectivity (with magnetic resonance imaging [MRI]). Topics will concentrate on the use of neuroimaging as a diagnostic marker and as a proxy of neuropathological staging. Additional topics to be addressed are the examinations of the Amyloid/Tau/Neurodegeneration (A/T/N) framework, the latest findings from temporal sequencing and propagation modeling studies, the use of neuroimaging to inform neurodegeneration and loss of synaptic integrity, advances in subcortical imaging, PET neuroimaging as a validation of fluid markers, and finally, neuroimaging in autosomal dominant AD. As the field evolves in the development and proliferation of in vivo neuroimaging markers of AD neuropathology, a parallel wave of meta-analytic approaches to neuroimaging cohorts is becoming the norm [3–6], with the focus also turning to machine learning approaches [7, 8] and data harmonization across cohorts [9]. Further, the increasing number of longitudinal cohorts with multimodal neuroimaging data (particularly those that are open to data sharing) have catalysed the publication of an unprecedented level of neuroimaging findings in Alzheimer’s disease across the preclinical and prodromal spectrum. Neuroimaging has contributed invaluable in vivo evidence towards neuropathological findings (i.e. validation), formed the primary mode of evaluating the effect of treatments on pathological load in clinical trials while fluid markers were still being refined, and provided unique and novel insights into the in vivo evolution of the disease that has helped to contextualize the inherently static/cross-sectional findings of neuropathology data. This review will cover the most recent and advanced sub-topics of AD neuroimaging with the intent of highlighting the continued importance of neuroimaging in the field, even with the advent of novel blood biomarkers. What remains the purview of neuroimaging is the in vivo topographical information that can both validate neuropathological literature, but also provide unparalleled access to the natural history of the AD pathological cascade during life.

Positron Emission Tomography Markers of Aβ and Tau

The focus of this review will be on more recently developed PET markers of Aβ and tau. The most common Aβ radiotracers are ¹¹C-Pittsburgh Compound-B (PiB) [10, 11], and the ¹⁸F compounds, such as Florbetapir [12, 13], Florbetaben [14, 15], Flutemetamol [16, 17], and Flutafuranol (also known as NAV4694) [18]. Currently, the Aβ-PET tracers approved for clinical use by regulatory bodies are Florbetapir [19], Florbetapen, and Flutemetamol [20]. Tau tracers are still being developed, with the most commonly reported tau-PET tracer being ¹⁸F-Flortaucipir (previously AV1451/T807); this tracer was recently approved by the Food and Drug Administration under the name Tauvid [21]. Newer tracers have also been developed, like ¹⁸F-MK6240 [22], ¹⁸F-RO948 [23], and ¹⁸F-PI2620 [24]. Comprehensive reviews on the state of tau tracers are already available and will not be touched upon here [25, 26].

Synergy Between PET Markers of Aβ and Tau

While post-mortem studies demonstrate that Aβ and tau proteinopathies emerge in vastly different regions, Aβ-PET and tau-PET signal are moderately correlated in both clinically unimpaired and impaired samples [27–33], particularly in medial temporal regions [31, 34, 35]. Indeed, Flortaucipir-PET exhibits a high sensitivity to distinguish between high and low Aβ-PET signal [36–38]. Clinically normal older adults with evidence of high Aβ exhibit higher tau-PET signal in medial and lateral temporal regions than those with low Aβ, as well as parietal and frontal regions [27, 33–35, 39–42]. Recent efforts to isolate the rhinal sulcus in Flortaucipir-PET images from individuals at various stages of impairment, as well as controls and E280A mutation carriers, have suggested that early signal in this region appears independently of Aβ-PET signal and that temporal neocortical tau-PET signal may emerge downstream as a result of increasing Aβ-PET abnormality [43]. Cross-sectional scatterplots between these two PET markers of Aβ and tau (MK6240 signal log transformed) in a sample of both clinically unimpaired and impaired individuals have suggested a sigmoidal relationship, with accelerating levels of tau-PET signal beginning ~ 25CL in the clinically normal group [44]. This paper also reported medial temporal high tau (T+; as defined by 95th percentile in Aβ- individuals) appearing first in the entorhinal cortex, followed by amygdala, hippocampus, and finally the parahippocampus in clinically normal individuals, supporting a sequential spread of tau in the medial temporal cortex. Further, highlighting the synergy between these proteinopathies to influence downstream events, clinically normal individuals with high levels of both Aβ-PET and tau-PET also exhibit the fastest rates of cognitive decline [32, 45–47]. Interestingly, in models examining the influence of Aβ-PET, tau-PET, and structural MRI (neurodegeneration) separately, tau-PET remains as the only marker associated with cognitive decline [46], suggesting its proximity—and specificity—to the clinical syndrome.

PET as a Marker of Neuropathological Staging

Neuropathological staging of AD suggests widespread neocortical distribution of Aβ plaques in the earliest stages of the disease, followed by involvement of subcortical and allocortical regions [48, 49]. Although many studies use a global cortical composite to indicate level of Aβ burden with PET, examination of regional Aβ-PET signal for staging may provide greater sensitivity for detection of early cognitive change (see Fantoni and colleagues [50] for a comprehensive review). For instance, cortical Aβ-PET signal has been found to appear prior to signal in the striatal region in sporadic cases of AD [50], although striatal and thalamic signal has been reported as an early marker in many AD mutation carriers ([51]; discussed later). A classification system that stratified both cortical and striatal Aβ-PET signal was found to better predict hippocampal atrophy and cognitive impairment across the clinical spectrum than cortical signal alone [52–54]. A four-stage model was also proposed in ADNI using Florbetapir-PET, which suggested regional signal first appearing in temporobasal and frontomedial areas, followed by the rest of the neocortex, the medial temporal lobe, and finally the striatum [55]. In addition, early Florbetapir-PET accumulation in posterior cortical regions has also been noted in clinically normal older adults [56], and in association with cognitive changes in initially low Aβ middle-aged individuals [57]. Lastly, ‘early’, ‘intermediate’, and ‘late’ regional composites of Aβ-PET accumulation have been developed [56]. For instance, ‘early’ accumulation sites were represented by the precuneus, posterior cingulate, isthmus cingulate, insula, and medial and lateral orbitofrontal cortices. Interestingly, gene expression associations were found to be different across the composites, suggesting selective vulnerability of brain regions to Aβ throughout the course of the disease. Taken together, latest regional amyloid neuroimaging research suggests that the earliest accumulation sites may exist in areas consisting of the temporobasal, posterior cortical, precuneus, posterior cingulate, isthmus, insula, medial, and lateral orbitofrontal region. Alternative perspectives, however, suggest that Aβ-PET may accrue simultaneously across the brain and that any regional deviation is due to heterogeneity in ‘carrying capacity’ (the maximum concentration of Aβ that can occur in any given region) [58]. The authors argued that the evidence for and against topographical sequences to Aβ accumulation may depend, to some degree, on assumptions placed on the data and the approaches to modeling Aβ-PET signal.

Although some heterogeneity exists in the literature regarding which regions show early, meaningful Aβ accumulation, findings for tau-PET seem much more consistent with the neuropathological findings. Flortaucipir-PET signal has been shown to track alongside Braak staging of tau tangle deposition [33, 37, 39, 41, 59], with elevated signal clearly delineating those with Braak tangle stage of IV or greater from those without at post-mortem [60]. For Flortaucipir-PET, methods have been proposed to identify topographical stages of signal, with ROI and voxel-based clustering approaches somewhat aligning with profiles consistent with Braak staging [33, 37, 59, 61], although perhaps appearing more widespread and spatially redistributed than suggested by the earliest Braak stages [61, 62]. Data-driven clustering approaches have also exhibited a relatively good overlap with Braak stages [61, 63]. Importantly, however, longitudinal tau-PET studies suggest that spatial tau accumulation patterns may be more heterogeneous than the Braak staging scheme might indicate no matter the clinical stage [64–66].

Longitudinal Changes in Tau-PET

Recent studies of longitudinal measures of Flortaucipir-PET suggest that frontal and temporal lobes show the fastest rates of change in sporadic AD [65, 66] and that baseline measures of tau-PET signal might align more closely with the clinical phenotype than slopes [65]. Tau accumulation rates in clinically unimpaired individuals with high Aβ-PET are relatively low [64, 66, 67], with high Aβ-PET being the strongest predictor of change [68, 69]. Other cohorts show a close association between higher baseline Flortaucipir-PET signal and faster accumulation rates [67]. In a study examining an 18-month period of follow-up, those with lower global Flortaucipir-PET signal showed the fastest rates of accumulation in the temporal lobe, while those with higher global signal showed faster accumulation in parietal and frontal regions [69]. Jack and colleagues [64] reported that clinically normal individuals with high Aβ-PET exhibited faster tau accumulation in basal and mid-temporal, retrosplenial, posterior cingulate, and entorhinal regions relative to those with low Aβ-PET, suggesting a widespread change in tau not limited to the medial temporal regions. Atrophy patterns in AD patients are also reportedly more widespread, with tau accumulation mirroring atrophy patterns in the same regions [66]. Cognitively impaired patients show faster rates of tau accumulation in cortical regions relative to medial temporal regions [64]. Higher rates of Flortaucipir-PET change have also associated with faster rates of cognitive decline [45], particularly in those who have high Aβ-PET [68] or faster cortical thinning [70]. In brief, tau-PET imaging seems to demonstrate signal that is proximal to neuropathological staging criteria despite issues related to off-target and non-specific binding. Some postmortem evidence does, however, suggest that there may be issues with attempting to quantify the very earliest tau changes with PET. For instance, examining cases of patients with Braak I-III NFT (i.e. low AD neuropathology change) was found not to be distinguishable from younger controls [71].

Off-Target Binding and Newer Tau PET Tracers

Flortaucipir-PET is associated with off-target binding in vessels, iron-associated regions, substantia nigra, calcifications in the choroid plexus [72], and neuromelanin and melanin-containing cells [73–75]. Due to strong off-target Flortaucipir-PET signal appearing in the choroid plexus adjacent to the hippocampal formation [27, 76], a commonly reported alternative in the literature is in the inferior temporal lobe, which has been associated with AD-related change [27]. Newer tau tracers, such as MK-6240, RO-948, and PI-2620, show greater affinity for tau deposits in AD brain tissue and less off-target binding in choroid plexus [77–79], although autoradiography validation suggests potential issues with binding to neuromelanin and melanin-containing cells [80] and retention in the skull/meninges [78] with the first two tracers. PI-2620 may show some off-target binding in the retina, substantia nigra, choroid plexus, and venous sinuses in healthy older adults [81]. Another issue is that non-AD tauopathies show tau-PET binding that is distinguishable from AD patients in signal strength but remains quantitatively higher than clinically normal controls [71] (although qualitatively the regional pattern of distribution may be dissimilar). Specifically, Flortaucipir-PET uptake in clinical syndromes and mutations associated with TDP-43 pathology have been noted [82]. As such, these cases are best described as ‘low affinity’ rather than ‘no affinity’ of the tracer to non-AD tauopathy [82]. This has also been seen to a lesser extent with other tau tracers like RO948 [83], although these are still preliminary findings. Tau tracer development continues to evolve with the aim of reducing off-target binding and improving sensitivity to tau tangle deposition.

Aβ and Tau-PET Signal Associations with Demographics

In clinically normal older adults, observational cohorts suggest that age and genetic risk for sporadic AD, apolipoprotein ε4 (APOEε4) status, are robust predictors of high Aβ-PET signal [84–86]. Findings from the screening cohort of the anti-amyloid treatment in asymptomatic Alzheimer’s disease (A4) Study suggest that clinically unimpaired individuals with high Aβ are older have a family history of dementia and are more likely to carry at least one APOEε4 allele (genetic risk for sporadic AD dementia) [87]. For tau-PET, age reveals interesting associations depending on the diagnostic group. Particularly in clinically normal individuals, older age is associated with higher signal [33, 59]. Cross-sectional associations between Flortaucipir-PET and age in the Mayo clinic cohort showed a moderate positive relationship across most ROIs, particularly in those who were clinically unimpaired [33]. Early and late onset AD patients, however, show negative tau-PET associations with age [88–90]. For atypical variants of AD, such as posterior cortical atrophy (PCA) and logopenic variant of primary progressive aphasia, younger age was strongly associated with higher tau-PET signal [89, 91], particularly in parietal regions [91]. An examination of cross-sectional MK6240-PET has revealed weak positive associations with age in clinically normal individuals and moderate negative associations with age in clinically impaired individuals with abnormal Aβ [44]. These negative associations in the clinically impaired have been rationalized as due to comorbidities. Older adults are more likely to present with clinical symptomatology due to multiple comorbidities beyond Aβ and tau, and as such, may require less AD-related pathology to induce compromised cognitive performance, unlike younger patients. For longitudinal Flortaucipir-PET, younger clinically impaired patients [64, 68, 92], as well as females [92], have been shown to exhibit faster rates of accumulation, further cementing an interesting relationship with age. Sex differences are also apparent, with clinically normal females showing higher signal in multiple regions relative to males, such as medial temporal, rostral middle frontal, and parietal regions [32, 93–95]. These findings are moderated by Aβ [93] and APOEε4 [96]. It is important to note those studies that do not find sex differences, cross-sectionally [97], or longitudinally [68]. Although some studies report sex differences in Aβ-PET [98, 99], and proximity to parental age at onset in sporadic AD may influence Aβ-PET signal in females [100], findings are not reported consistently perhaps due to differences in age between these cohorts [101]. As such, the literature suggests age and APOEε4 status are risk factors for elevated signal in PET markers of both Aβ and tau, with sex showing divergent patterns.

Aβ and Tau-PET Signal Associations with APOE

While the relationship between cross-sectional Aβ-PET and APOEε4 is well documented, more recent studies have examined the natural history of Aβ accumulation in APOEε4 carriers [104,105,106]. Clinically normal older APOEε4 carriers with high Aβ-PET signal exhibit faster rates of cognitive decline than non-carriers [6, 102, 103]. Carriers also show earlier signs of accumulation in Aβ-PET than non-carriers, particularly if they are homozygotes [104, 105]; this is true even in carriers with initially normal levels of Aβ-PET signal [106]. The prevalence of APOEε4 carrier status is much higher in those with abnormal Aβ (PET or CSF), with approximate prevalence estimates of 51% and 65% in clinically normal and MCI/AD dementia patients, respectively [107]. Higher tau-PET signal has also been shown in APOEε4 carriers [40], isolated to the medial temporal region [40, 89, 108, 109]. A recent study using MK6240-PET suggests higher signal in the entorhinal cortex and hippocampus in APOEε4 carriers that was independent of Aβ-PET signal, sex, age, and clinical diagnosis [109]. In clinically impaired individuals who have abnormal Aβ-PET, higher neocortical Flortaucipir-PET signal has been reported in non-carriers [108, 110], particularly in parietal regions, and a stronger relationship between Flortaucipir-PET and cortical thinning relative to APOEε4 carriers [108]. Reasons for this association remain unclear but highlight the potential for different APOEε4 pathways for tau propagation. Recent findings using longitudinal Flortaucipir-PET have also suggested that APOEε4 carriers show faster rates of accumulation in a combined sample of clinically unimpaired and impaired participants [68], although this has not been supported in other cohorts [67]. Findings with APOEε4 suggest interesting disease-dependent and independent pathways associated with each proteinopathy.

Neuroimaging and Ethnoracial Factors

There are very few studies assessing ethnoracial differences in AD neuroimaging markers and remain an area requiring more focus [111, 112]. This is largely due to disparities in the inclusion of non-white populations in biomarker studies [111, 113], but also include a wide variety of other challenges that are covered in detail in previous white papers [111]. With regard to PET markers of Aβ and tau, higher levels of Aβ-PET signal have been reported in African Americans relative to Caucasians, after adjusting for age, sex, APOEε4, and CVD [114], although this was not supported in a sample predominantly representing those of Caucasian descent [115]. Further, abnormal levels of Aβ-PET in African American APOEε4 carriers has been associated with faster decline in executive function/speed [116]. Bearing this in mind, the literature to date in other sections of this review focus predominantly on findings from cohorts that are typically low in diversity, and as such, should be considered with an appropriate caveat.

A/T/N Framework

Arising, in part, from a hypothetical model of dynamic changes in AD biomarkers [2], the A/T/N framework was proposed [1]. This framework details a data-driven classification scheme for AD biomarkers, which are binarized within each of the three classes: ‘A’ to denote amyloid, ‘T’ to denote tau, and ‘N’ to denote neurodegeneration. For the purposes of this review, we will refer to papers that have probed the A/T/N framework. The ‘N’ category can be defined as a range of different measures, such as hippocampal volume, cortical thickness, FDG-PET signal, or a combination of these [1, 117]. Early work using A/N reported that rates of brain atrophy were highest amongst A+/N+ [118] and that higher rates of Aβ-PET accumulation were not contingent upon lower hippocampal volume [119]. Other work examining the synergistic nature of A and N have reported the impact of the two on poorer cross-sectional memory performance [120] and faster rates of cognitive decline in cognitively unimpaired older adults [6]. For A/T profiles, using previously established cut-offs for Aβ and tau, Jack and colleagues [121] found that participants predominantly represented three groups: A−/T−, A+/T−, and A+/T+ with the A−/T+ group considered incredibly rare. By contrast, Weigand and colleagues [122] reported that A−/ + comprised 45% of the sample compared with only 6% in the A+/T− group. A−/T+ showed moderate decreases in cognition, with the authors arguing that findings were discordant with the ‘amyloid first’ hypothesis. Mattsson-Carlgren [123] reported very different proportions of A/T groups again with their binarization suggesting different interpretations that can arise in the A/T framework. In a separate study, clinically normal adults who were A+/T+ exhibited the worst cognitive outcomes and higher rates of progression to MCI, particularly at older ages [47]. Papers examining A/T/N have found that APOEε4 carriers are twice as frequent in A+ groups and that the A+/T+/N+ classification is the most prevalent in those over 80 years [124]; by the time individuals are 85 years, a majority exhibit at least one A/T/N biomarker. In one study, tau was found to be higher in A+/T−/N− individuals than all-negative controls, while greater N was found in the A+/T+/N− vs all-negative controls, suggesting a temporal sequence of biomarker abnormality [125]. A recent study also reported that Aβ-PET was consistently apparent in large spatial areas of grey matter in AD patients, followed by tau-PET, and then MRI/FDG-PET, with approximately 34% of all grey matter voxels exhibiting signal for all three markers (compared with ~ 1% for clinically normal individuals) [126]. Taken together, the literature suggests that A/T/N can provide some useful heuristics for measuring progression of disease. There remains some disagreement as to the exact definition of groups, specifically in the approach to dichotomizing tau-PET, which will require further exploration with future iterations of the A/T/N framework. Specifically, the binarization of non-bimodal distributions, such as for tau-PET and MRI measures of hippocampal volume, seems to result in different ATN proportions depending on the method and also has very largely implications on the estimation of future cognitive decline between the groups [123]. As such, further clarity and standardization need to be implemented in this area for meaningful conclusions to be drawn.

Temporal Sequencing and Propagation Modelling

Seminal work from autosomal dominant Alzheimer’s disease (ADAD) [127] was some of the first to recapitulate the well-known ‘Jack curves’ that depict a hypothetical model of biomarker and cognitive change along the clinical spectrum of AD [2]. A cascade of changes, starting with Aβ-PET accumulation, were followed by glucose hypometabolism and hippocampal volume shrinkage [127]. Using event-based modelling to estimate cross-sectional biomarker sequence change in the same ADAD cohort, Oxtoby and colleagues [128] reported changes in cortical then subcortical Ab-PET signal, followed by changes in CSF p-tau₁₈₁, total tau, and Ab, then atrophy in putamen and nucleus accumbens, and finally cognitive decline and cortical glucose hypometabolism. In sporadic AD, the same methods were employed on cross-sectional data across 2389 clinically normal and impaired individuals [129]. The authors found the first biomarkers to become abnormal were CSF markers, followed by cognitive measures and then volumetric measures later in the disease stages. Interestingly, slightly different methods resulted in divergent sequences of events. Change-point timescales have been developed using cross-sectional CSF and MRI data that suggest a sequence where brain atrophy emerges downstream to CSF Aβ and tau abnormality [130]. Other cross-sectional sequencing depicts a pattern of rising abnormality in neocortical Aβ-PET, followed by focal regional hotspots of tau-PET, and finally changes volumetric MRI restricted to the temporal lobe [41]. Although earlier work relied on cross-sectional data, studies using longitudinal data in sporadic AD are now shedding light on associated changes between these in vivo neuroimaging markers. Latent time joint mixed-effects methods have been applied to multimodal longitudinal neuroimaging and cognitive data [131] to uncover dynamic change akin to the Jack curves [2]. Insel and colleagues [132] extended this method to create an ‘estimated disease time’ akin to the estimated years to onset (EYO) reported in ADAD mutation carriers [127]. In this study, early changes in Aβ-PET were observed in the posterior cingulate and precuneus, with tau-PET signal found to initially change in the medial and lateral regions of the temporal lobe [132]. Koscik, Betthauser, and colleagues [133] proposed ‘amyloid chronicity’ metric, a timescale based on the age at which a cognitively normal individual’s Aβ trajectory is likely to hit positivity. Those who were proposed to sit later on the chronicity scale exhibited faster cognitive decline and higher entorhinal tau-PET signal. Other studies predict changes in one neuroimaging modality relative to another. Leal and colleagues [134] found that Aβ-PET accumulation rates were associated with higher Flortaucipir-PET signal at a later timepoint in clinically normal individuals who were initially low Aβ. These results provide supporting evidence of a downstream change in tau signalled by accumulating Aβ levels. Evidence suggests that Aβ-PET accumulation is associated with subsequent changes in tau-PET, which in turn are associated with cognitive change [135].

Recent data-driven models of tau propagation align with the neuropathological staging of the disease [136–138]. Data reduction of voxel-wise Flortaucipir-PET signal suggested ten independent tau networks that moderately overlap with known intrinsic fcMRI networks [137]. A joint independent component analysis of both Flutemetamol-PET and Flortaucipir-PET images across individuals of various levels of cognitive impairment resulted in nine factors of orthogonal patterns of Aβ and tau [139]. These were found to have moderate overlap with functional networks from functional connectivity data [139], contributing evidence in support of the seminal study by Jones and colleagues [140] showing spatial patterns of pathological deposition contributing to widespread and ‘cascading’ network failure. Specifically, one propagation model ‘seeded’ tau in different regions of the brain and simulated diffusion through functional and anatomical connections [136]. Findings suggested that tau may spread from the entorhinal cortex, through the amygdala, hippocampus, parahippocampus, and fusiform, followed by lateral temporal regions and finally into other areas of the cortex. Extending this argument of early tau seeding in the entorhinal cortex, Adams and colleagues [141] examined the functional connectivity patterns in sub-regions of the entorhinal cortex in clinically normal individuals and found that connectivity in the anterolateral regions was more closely associated with tau-PET signal in connected regions, than in posteromedial regions. Findings here supported the impact of early transentorhinal tau deposition on downstream appearance of tau on functionally connected regions. This study, however, was conducted cross-sectionally and was limited in predictive scope. A different approach using longitudinal data from clinically normal individuals with high Aβ showed that regions with faster rates of tau accumulation were more likely intrinsically connected with other regions that had faster rates of accumulation [5]. This mirrored cross-sectional data from the same group showing that inferior temporal tau-PET signal was associated with tau-PET signal in functionally connected regions of interest, and independent of Aβ-PET burden or clinical symptomatology [142]. A growing number of studies are also investigating the notion of ‘epicentres’ in the brain that may harbour early pathological burden and serve as the vehicle for propagation of Aβ and tau species. In semantic variant primary progressive aphasia and frontotemporal dementia patients, a recent study showed that patient-specific epicentres can be identified and placed within a template of a functional human connectome [143]. These epicentres can act as predictors of longitudinal brain atrophy within-individuals through functional connections based on graph theory metrics. Studies involving atypical variants of AD (posterior cortical atrophy and logopenic variant progressive aphasia) have also shown ‘epicentres’ of high Aβ-PET signal and high tau-PET signal may show differing associations with functional connectivity metrics, perhaps highlighting different mechanisms for proteinopathy propagation. Finally, in two independent cohorts of clinically normal adults, and patients with mild cognitive impairment and AD dementia, Franzmeier and colleagues [61] found that seed-based connectivity in tau epicentres (defined as regions that exhibited greater likelihood to have high tau positivity based on Gaussian mixture modelling applied to Flortaucipir-PET images) were associated with normative connectivity patterns, suggesting that tau pathology spreads through connected, not necessarily adjacent, regions. Interestingly, there was some level of heterogeneity in spatial maps of tau-PET signal, which seemed to follow the tau epicentre connection model. Tau-PET covariance patterns in AD patients with intrinsic connectivity template maps from healthy young adults also show striking convergence [144]; Aβ-PET covariance patterns are far less cohesive. Shokouhi and colleagues [145] proposed that functional network characteristics in females may support faster tau propagation relative to males, suggesting one rationale for sex differences in tau-PET signal.

With regard to Aβ progression models, regional Aβ progression data may be less clear due to the properties of the Aβ-PET signal [58]; however, there is evidence to suggest early sites of Aβ-PET deposition. Early work using epidemic spreading models on Aβ-PET from Alzheimer’s disease Neuroimaging Initiative (ADNI) was able to reproduce Aβ patterns seen across the clinical trajectory, although the model could only explain ~ 50% of all variation in Aβ-PET deposition. Specifically, the model identified the cingulate cortices as the most likely starting seed regions for Aβ, similar to aforementioned work [132]. Using longitudinal Florbetapir-PET images and cross-sectional region-specific thresholds in clinically normal older adults, Jelistratova and colleagues [146] found that predicted regional Aβ staging was found to be congruent across both cross-sectional and longitudinal data and in approximately 98% of individual Aβ topographical patterns. They reported early accumulation in anterior temporal regions, followed by frontal and medial parietal regions and associative neocortex. In a very large sample (n = 3327) across a range of diagnoses, a pooled multi-tracer regional Aβ threshold ranking method found Aβ abnormality to initially appear in cingulate, followed by orbitofrontal, precuneus, and insular cortices and finally associative, temporal, and occipital regions [147]. Similar staging patterns were describe in an earlier sections of this review [56]. What has been consistently shown is the characteristic ‘sigmoid’ accumulation curve of Aβ-PET over time which is formed by integrating the change signal relative to baseline Aβ-PET signal [134, 148, 149]. In addition, approximate time-courses for Aβ accumulation from a status of low to ‘high’ (a threshold of abnormality) Aβ is suggested to occur over approximately 15–17 years [148, 149]. The notion of a threshold of abnormality for Aβ-PET, traditionally defined using Gaussian mixture models or other statistical or visual approach, may be overly conservative or stringent relative to neuropathological markers. Evidence supports that a far lower threshold of Aβ-PET is associated with neuropathological evidence of Aβ at post-mortem [3, 150], and which can inform those who are likely to show faster rates of cognitive decline and Aβ accumulation over time [151].

An important caveat to keep in mind is that results from many of these studies are largely modelled on cross-sectional data, and so it is important to keep in mind the caveat that these findings will need to be replicated with longitudinal data. Further, these methods used often involve machine learning/AI and simulations of the data, which require strong modelling assumptions to be met. Recent reviews of these approaches have strongly advocated for rigorous external validation and replication, ideally by independent investigators, before the models can be considered reliable [152, 153]. And finally, the risk of overfitting these models is also very high, even if cross-validation is employed, particularly if parameters are selected on the basis of the best cross-validated measure of performance (‘overhyping’) [154]. Given all these caveats, it is important to acknowledge that findings from current propagation modelling approaches must be consolidated and interpreted with caution as per the performance and claims of these methods.

Machine Learning/Prediction

There is a rapidly growing and exciting literature involving the use of machine learning to make out of sample predictions about clinical progression in AD, which has culminated in The Alzheimer's Disease Prediction Of Longitudinal Evolution (TADPOLE) challenge [8]. The aim of these approaches, and TADPOLE specifically, is to train models and algorithms on ADNI and other open-source data to make monthly predictions on outcomes for participants (clinical diagnosis, global cognition and total ventricle volume). Application of state-based approaches such as minimal recurrent neural networks and linear state space models has shown promise as they can handle missing data, have shown some promise with individuals who only have one timepoint, and can, in theory, forecast outcomes quite far into the future [155]. Other successful approaches have implemented non-deep-learning algorithms and other features in their models [156]. The best approach so far for ventricle prediction has been from team EMC1, which used disease progression modelling and spline regression [7]. This area has great translational potential given the ability to be cross-validated and forecast unseen data, rather than just fit to pre-existing data as is commonly done. This area still requires a lot of refinement and investigation but deserves mention in this review.

Neurodegeneration and Loss of Synaptic Integrity and Connectivity

Neurodegeneration and loss of synaptic integrity is a characteristic component of AD; however, it is not solely driven by AD processes [157] and is non-specific for the disease [1, 119]. Hippocampal volume is also predicted by multiple neuropathological contributors, such as TDP-43 and hippocampal sclerosis [157]. Jacobs and colleagues [158] suggested that the strongest association for memory performance in the hippocampal formation was with presubiculum volume; this relationship remained even after adjusting for Aβ and tau-PET signal, highlighting the importance of regional hippocampal specificity in relation to the earliest preclinical signs of neurodegeneration in AD. Further supporting the regional importance of the hippocampal formation, in vivo volume of the posterior hippocampal regions are more closely associated with tau tangles at post-mortem, while in vivo volume in anterior regions and entorhinal cortices may be more closely associated with TDP-43 neuropathology [159]. Indeed, faster rates of atrophy in medial temporal regions (i.e. amygdala, entorhinal cortex, hippocampus) are associated with more advanced stages of TDP-43 load [160], demonstrating the unique contribution of TDP-43 pathology to neurodegeneration. Further highlighting the lack of specificity of neurodegeneration to AD, a study that implemented data reduction methods to structural MRI data found no morphological networks that discriminated AD patients from older or younger adults, although multiple networks were found to discriminate based on age alone [161].

Although MRI markers of brain atrophy may be non-specific, they form an inherent downstream consequence of Aβ and tau abnormality. Specifically, the medial temporal structure has remained of most interest in the field, and a comprehensive review on hippocampal subfields and AD has recently been published [162]. Evidence is also building from both cross-sectional and longitudinal data to suggest that the association between tau-PET signal and brain atrophy is most likely found in those who are already Aβ-PET positive [36]. In clinically normal individuals, baseline inferior temporal tau-PET signal is associated with prospective cortical thinning in temporal, midline, and prefrontal regions [70]. For patient populations, the association between tau-PET signal, cortical thinning, and regional grey matter volume is much stronger than for clinically normal cohorts [163, 164]. La Joie and colleagues [164] showed that the global intensity of tau-PET, but not Aβ-PET, predicted the rate of subsequent cortical thinning in a cohort of patients, which was independent of baseline thickness. Similar findings have been reported cross-sectionally [165]. Further, the spatial pattern of tau-PET signal is associated with the topography of future atrophy, at the individual-level, for AD dementia patient (particularly younger onset) [164], suggesting the proximal nature of tau and atrophic events. Regional tau-PET signal is considered to be more closely associated to markers of synaptic dysfunction [40], neurodegeneration [41, 166], and cognitive impairment [41, 45, 59, 167], than markers of Aβ. Evidence is also points to both synergistic and independent influences of cortical thinning and tau-PET burden on cognitive performance [167, 168].

Structural connectivity is becoming increasingly of interest in preclinical AD, with studies suggesting grey matter structural networks become disrupted in AD [169, 170]. Disruption of these networks in clinically unimpaired older adults is associated with faster rates of progression to MCI or dementia [171], and levels of Aβ burden [172]. Recent findings also suggest that loss of integrity of grey matter networks presages hippocampal atrophy in those with high Aβ [173]. Decline in network measures have also been associated with contemporaneous decline on cognitive tasks; in controls, for instance, this association was strongest with decline in the memory and language domains [174]. Graph theory applied to diffusion tensor imaging has also revealed longer paths, lower efficiency, and great modularity of structural networks in clinically normal older adults who are A−/N+ and A+/N+ , with this loss of organization impacting memory performance in A+/N+ individuals [175]. These findings highlight the co-dependence of neurodegeneration on structural integrity, and also suggest, to some degree, of synergy between amyloid and neurodegeneration to influence progressive disruption of structural networks. Finally, recent work examining global white matter diffusion characteristics has supported independent contributions to cognitive decline above and beyond the influence of amyloid [176], emphasizing the importance of white matter degradation in the clinicopathological pathway. In autosomal dominant AD mutation carriers, mean diffusivity in posterior parietal and medial frontal regions has been shown to be higher than in non-carriers, and associated with lower grey matter volume, lower CSF Aβ, higher CSF p-tau₁₈₁, and soluble CSF TREM2 [177]. It was also projected to appear as early as 5–10 years prior to estimated symptom onset, suggesting very early degradation perhaps arising from AD pathology and microglial activity.

With regard to white matter hyperintensities (WMH), Aβ-PET-associated WMH were found in periventricular regions (frontoparietal regions) in clinically unimpaired individuals, which associated with microbleeds, suggesting a contribution of cerebral amyloid angiopathy to this relationship [178]. By contrast in this same study, tau-PET was not associated with WMH. In a separate cohort of clinically unimpaired adults with a wider age span (30–89 years), Aβ-PET relationships with WMH were found to be moderated by age, both with global WMH, but also posterior parietal regions [179]. Some studies have not shown a cross-sectional association between WMH and Aβ-PET [180], but rather global, parietal, and frontal WMH related to Aβ-PET accumulation in clinically normal older adults who were initially exhibited low Aβ-PET.

Changes in intrinsic functional connectivity in cognitive networks, particularly the default network, may also represent an informative marker of encroaching Alzheimer’s disease. Longitudinal examinations of connectivity integrity have suggested that posterior default mode network failure might appear prior to the onset of abnormal Aβ-PET load and that hyperconnectivity between this sub-network hub and other nodes (predominantly frontal) may be associated with contemporaneous Aβ-PET accumulation [140]. A model of ‘cascading network failure’ has been proposed that suggests an overload of functional activity triggered by higher Aβ burden, which spawns widespread system failures [140]. More recently, clinically unimpaired individuals with high Aβ-PET were shown to exhibit decreased intrinsic connectivity between the medial temporal lobe and regions in the anterior-temporal systems [181]. With the onset of clinical symptoms, this same study showed reduced connectivity between the medial temporal lobe and posterior-medial regions, highlighting the fragility of intrinsically connected networks with increasing severity of AD clinicopathology [181]. Earlier work suggested high Aβ-PET was localized to regions of cortical functional hubs [182] and that elevated Aβ-PET was associated with lower functional connectivity at the cross-section, particularly in the default mode network [182–185], but these findings have not been consistently replicated [186]. Studies using graph metrics have also suggested that those with CSF evidence of preclinical AD exhibit compromised clustering coefficients and modularity relative to controls [187]. Regardless of the connection between Aβ-PET and fcMRI at the cross-section, there does appear to be a synergy between Aβ and intrinsic connectivity influencing cognitive decline, suggesting perhaps a unique contribution of both to influence downstream clinical outcomes. Functional connectivity is more strongly associated with tau-PET signal than with Aβ-PET [142]. Indeed, Aβ and tau-PET are associated with an interesting pattern of intrinsic connectivity in the default mode and salience networks of clinically normal individuals [188], with high Aβ associated with hyperconnectivity when tau-PET signal is low, but hypoconnectivity when tau-PET signal is high. Taken together, these findings highlight the cascading loss of synaptic integrity with disease progression and the potential influence of AD biomarkers on this propagating network failure and downstream cognitive loss.

Imaging Blood–Brain Barrier Permeability and Perivascular Space

There have been major advances into investigations of blood–brain barrier permeability using dynamic-contrast-enhanced MRI [189–191]. Initial work suggested that blood–brain barrier permeability (measured with a K_trans constant using post-processing analysis of these images) showed age-dependent vulnerability in the hippocampal region, which was also associated with CSF markers of blood–brain barrier-related pericytes [191]. More recently, Montagne and colleagues reported that permeability of the blood–brain barrier was the greatest in APOEe4 carriers and that this permeability was increasingly apparent with as cognitive impairment became more severe [189]. In addition, baseline levels of blood–brain barrier pericyte injury markers in CSF predicted future cognitive decline in APOEe4 carriers relative to non-carriers, highlighting one mechanism explaining vulnerability to AD associated with APOEe4. Another study assessing vascular dysfunction examined cerebral blood flow (using 3D pseudocontinuous arterial spin labelling), Flortaucipir-PET signal, and CSF measures of pericyte injury and found negative associations between cerebral blood flow and tau-PET signal specifically in temporo-parietal regions [192]. Further, these relationships, and associations between CSF markers and tau, were strengthened in those with greater Aβ-PET signal. Other advances in the field of vascular imaging and AD have been covered in great detail in the following reviews [193, 194]. Novel quantitative imaging approaches have also been developed for measuring perivascular space using MRI [195], which has traditionally been scored based on visual reading of the visibility of the space in certain brain regions. This new approach taps into the contrast of the image by combining T1- and T2-weighted images to remove non-structured high-frequency spatial noise from the image and allow for quantification of perivascular space. The technique was recently applied in the context of AD to show that increased nonparenchymal white matter fluid is observed around anterior thalamic radiation and sagittal striatum regions in cognitively impaired individuals [196] and may be the source of greater mean diffusivity in prodromal AD. Novel MRI techniques such as these have the potential to harness more information from structural imaging for the purposes of predicting onset of disease.

Subcortical Imaging

Neuroimaging of the basal forebrain, locus coeruleus, and other subcortical regions has gained increasing interest as sites that may accrue critical and early pathological insult in AD, supported by findings in neuropathology [197–199]. Cholinergic neurons of the basal forebrain are responsible for providing cholinergic input throughout the cortex, and rates of basal forebrain atrophy are reportedly faster in cognitively normal older individuals than whole brain atrophy [200]. The nucleus basalis of Meynert, which forms a core component of the basal forebrain, exhibits atrophy patterns concentrated in posterior regions of this cholinergic innervation in mild AD patients [201]. Basal forebrain volume has been associated with level of Aβ-PET signal at the cross-section [202–204], and in those with high Aβ-PET, longitudinal changes in nucleus basalis of Meynert have been reported to predict entorhinal volume and not the reverse [199, 205]. This directionality supports the hypothesis that selective vulnerability to the basal forebrain cholinergic projection system may be a consequential upstream event of entorhinal and neocortical degeneration and should be given more consideration in pathological models of AD.

The locus coeruleus has also been implicated early in the disease, with Braak’s work highlighting the presence of tau tangles in this region as some of the earliest in the pathological cascade [48]. Robust neuroimaging approaches are now being developed to examine this small brain region [206]; MRI contrast of the locus coeruleus is reduced in AD dementia patients and is associated with abnormal CSF Aβ—but not total tau—levels [207]. In addition, a more ‘youth-like’ locus coeruleus in older adults, measured with MRI contrast, is associated with better memory performance [208], demonstrating the importance of this noradrenergic region in cognitive maintenance. Probabilistic atlases that define tracts connecting the locus coeruleus to entorhinal pathway using high-resolution Human Connectome Project data have revealed that degraded fibre integrity in this pathway is associated with increasing clinical AD dementia severity [209]. With approaches still being developed to appropriately image the locus coeruleus [210], this will be a rapidly expanding area of research, particularly in preclinical AD.

PET Neuroimaging: Association with Fluid Markers and Neuropathology

PET neuroimaging has often formed a method of validation, both from the perspective of novel fluid markers and neuropathological studies. In this section, focus will be dedicated to novel plasma markers, cerebrospinal fluid markers, and finally to the latest neuropathological evidence associating with PET markers.

Plasma markers. The past few years has seen the development and validation of novel assays for plasma markers of Aβ and tau species. PET has formed part of the validation by which these measures are compared. Assays for plasma markers of Aβ show relatively good areas under the curve (AUC) for predicting abnormal Aβ-PET status [211–214], and tau-PET signal in AD-related regions [214]. High-performance plasma markers of phosphorylated-tau₁₈₁ (p-tau₁₈₁) show significant associations with tau PET signal [215–218]. Relationships between plasma p-tau₁₈₁ and Aβ-PET are less clear, with some showing a relationship [215, 216, 218] and others not [217]. In older adults from ADNI, baseline plasma p-tau₁₈₁ levels were associated with Flortaucipir-PET signal 6 years later in the medial temporal lobe and posterior cingulate in clinically normal older adults[219], and more widespread regions of the temporoparietal cortices in clinically impaired patients. Changes in p-tau₁₈₁ were more closely associated with Flortaucipir-PET signal, particularly in clinically normal older adults. Spatial distributions of association overlapped with topographical neurofibrillary tangle distribution [219]. Finally, changes in plasma p-tau₁₈₁ were reported to reach abnormal levels approximately 17 years after Aβ-PET abnormality. In another cross-sectional study, plasma p-tau₁₈₁ is reported to be predictive of high Flortaucipir-PET status [217], although CSF markers of p-tau₁₈₁ show stronger associations with Flortaucipir-PET signal [220]. Some papers have recommended plasma t-tau/Aβ_1-42 ratios to predict abnormal tau-PET status [221]. Findings with plasma markers of tau phosphorylated at threonine 217 (p-tau₂₁₇) show higher levels in clinically normal older adults with abnormal Aβ-PET and normal tau-PET versus those with both normal Aβ and tau-PET [222]. Event-based modelling suggested p-tau₂₁₇ abnormality occurred much earlier than entorhinal cortical tau-PET signal and that tau-PET accumulation evident only in those with abnormal levels of p-tau₂₁₇ [222]. These findings suggest that abnormal plasma markers of p-tau₂₁₇ could be the harbinger of downstream aggregation of insoluble tau tangles that are measured with tau-PET. Other studies of plasma p-tau₂₁₇ also suggests the discriminative accuracy of this marker to detect abnormal tau-PET is very high (area under the curve = 0.93). Finally, evidence suggests that those who are low Aβ-PET but have abnormal plasma Aβ₄₂/Aβ₄₀ at baseline are 15 times more likely to become Aβ-PET positive over 18 months [212]. As such, there seems to be clear associations between plasma and PET markers of Aβ and tau, with plasma markers appearing to become abnormal prior to PET. This field is rapidly expanding, and new information will continue to appear on these novel plasma markers and their associations with PET tracers.

Cerebrospinal fluid markers. With regard to CSF markers of Aβ and tau, Flortaucipir-PET has shown moderate relationships with p-tau and t-tau in clinically normal older adults [38, 223] and in clinical populations [39, 224–226]. Comparative to CSF markers, evidence suggests that Flortaucipir-PET signal may be more robust to detecting the more severe stages of AD dementia diagnoses [39, 227, 228] and may be more closely associated with brain atrophy and cognitive impairment in the latter stages of clinical impairment [36, 225]. One argument is that CSF t-tau and p-tau behave as markers of ‘disease state’ (elevation in those destined to be diagnosed with AD dementia, and prior to the emergence of tau tangles), while Flortaucipir-PET signal may serve as a better proxy for ‘disease stage’ (increases with stages of the clinical syndrome and is more proximal to stage-appropriate atrophy and cognitive change) [225]. Certainly, studies have shown that CSF and PET cannot simply stand in place of one another [123]; discordance metrics between CSF and PET markers have been examined for their implications [229, 230]. Aβ discordant individuals that are CSF+/PET− show faster rates of Aβ-PET accumulation and progression to CSF+/PET+ over 5 years [230]. With lower levels of CSF tau and Flortaucipir-PET after 5 years than CSF+/PET+ concordant individuals, the authors argued that this CSF+/PET− group may have a more favourable prognosis than those who were CSF+/PET+. Interestingly, CSF−/PET+ may have been classified largely due to processing error or non-specific binding in the white matter [230]. Another study suggested that for individuals showing CSF+/PET+ markers of tau, the likelihood of exhibiting an MCI/AD dementia diagnosis was high (76%), along with guaranteed amyloid positivity [229]. The proportion of those with CSF+/PET− was much greater than CSF−/PET+, again supporting the argument that CSF markers signal earlier change than PET. Barthélemy and colleagues [231] examined a variety of hyperphosphorylation sites for CSF tau and their changes over time. With regard to PET markers, CSF p-tau₂₁₇ had the strongest area under the curve for classifying abnormal Aβ-PET, followed closely by p-tau₁₈₁, indicating that Aβ-PET abnormality was proximal, or even after, phosphorylation events had taken place at these sites. Further, when examining quartiles of Aβ-PET signal, all sites exhibited a dose–response increase, except for pS202, which showed a decrease with increasing Aβ-PET levels. In this same paper, cortical thinning and subcortical thickness was reported to be most closely associated with p-tau₂₀₅ and p-tau₂₁₇ sites, suggesting that rise in the former may be more representative of cortical atrophy. Future examination of phosphorylation sites and their association with in vivo regional proteino-pathological deposition will be very critical for learning more about how these molecular changes are marking the stages of disease.

Neuropathological evidence. Seminal work for Aβ-PET validation using neuropathology is well established [3, 150, 232, 233]; the literature shows, across the commonly used tracers, that global Aβ-PET signal can differentiate neuropathological classifications of Aβ plaque burden. Further, neuropathologically defined thresholds for Aβ-PET burden are much lower than are attained using data driven approaches [3, 150]. Of interest for this review, more recent papers are now available on the association between tau-PET and neuropathology. Using phosphorylated tau (AT8) immunohistochemistry and Gallyas staining, one PSEN1 severe early-onset case study showed robust alignment between Flortaucipir-PET signal and density of tau-positive neurites and total tau burden at autopsy [234]. Another recent pilot study of three patients (two with AD dementia) showed that mean cortical Flortaucipir-PET signal during life was strongly associated with AT8 immunohistochemistry of brain tissue at death [235]. No correlation in subcortical regions was found, supporting the notion of ‘off-target’ binding in these regions in clinically normal older adults [74]. A sample of 67 valid autopsies of confirmed AD patients also showed high sensitivity and moderate specificity of Flortaucipir-PET to predict Braak stage V and VI [236]. In addition, an examination of a variety of patient groups by Soleimani-Meigooni and colleagues [71] recently showed that patients with non-Alzheimer tauopathies had lower Flortaucipir signal than AD patients but was still elevated compared with controls, suggesting that non-Alzheimer tauopathies may not be easily differentiable from AD-related tau. Taken together, these findings both validate these PET markers but also provide a cautionary tale for Flortaucipir-PET signal with regard to off-target binding and lack of specificity with non-AD tauopathy binding.

Neuroimaging in Early Onset Versus Late Onset AD

Heterogeneity exists in the clinicopathological expression of sporadic AD [237], which can be somewhat better understood when comparing early onset (EOAD) cases, often defined as those diagnosed prior to age 65, from late onset (LOAD) cases. In EOAD patients, prominent patterns of grey matter atrophy can be found in the posterior cingulate cortex [238] and temporoparietal regions [239], rather than the medial temporal lobe pattern demonstrated in LOAD. EOAD patients with posterior cortical atrophy are also more likely to display different patterns of hippocampal subfield atrophy, with greater atrophy in the subiculum and dentate gyrus whereas typical AD shows atrophy in the left CA1 and hippocampal tail [240]. EOAD patients also show stronger tau-PET signal than LOAD [29, 31, 40, 90], and greater levels of atrophy [239], given equitable levels of global cognitive impairment (although nonmemory domains are more likely to be affected in EOAD [237]) suggesting a more aggressive pathology in younger cases. In addition to showing stronger tau-PET signal, EOAD may also reflect different topographical patterns of tau-PET signal throughout the neocortex [40, 88, 90]. Neuropathological evidence supports this divergent pattern of tauopathy, with EOAD patients much more likely to display a hippocampal sparing pattern of neurofibrillary tangle deposition as compared with LOAD, which tends to show a limbic-predominant pattern of tangle burden [241]. The ‘hippocampal sparing’ AD subtype has also been revealed in vivo in AD patients using MRI algorithms and has been associated with AD-like hypometabolism on FDG-PET, poorer executive functioning and faster clinical progression than the ‘limbic predominant’ AD subtype [242]. Finally, recent evidence also suggests decreased functional connectivity within the default mode and the limbic networks in EOAD relative to LOAD, while EOAD showed lower FC in the frontoparietal and visual networks [243]. Taken together, these findings suggest a more aggressive pathological process in EOAD that is also represented by a different topographical pattern of pathological burden relative to LOAD.

Neuroimaging in Autosomal Dominant AD

Much can be elucidated about the Alzheimer’s disease pathological cascade from mutation carrier cohorts, particularly given the relative lack of confound with comorbidities in these younger individuals. ADAD cases from the Dominantly Inherited Alzhiemer Network (DIAN) [127] are reported to show elevated Aβ-PET signal up to 15 years before EYO, and cortical glucose hypometabolism up to 10 years prior, and cortical thinning in the medial and lateral parietal lobe approximately 5 years prior [244]. Longitudinal data has now supported these findings with rates of Aβ-PET accumulation increasing in mutation carriers approximately 19 years before EYO, followed by hypometabolism (14 years) and faster rates of brain atrophy (5 years) [245]. Aβ-PET signal in the precuneus is the first region to show divergence from non-carriers, highlighting regional Aβ pathological abnormality in mutation carriers that is not seen in preclinical sporadic cases. Other cohorts of ADAD have shown similar patterns of change [246]. In the Colombian Kindred, a cohort of descendants with a mutation in presenilin 1 (PSEN1) E280A [247], higher Aβ-PET signal in the striatum is associated with higher Flortaucipir-PET signal and memory decline [248]. Increased signal in this region does not correspond with overlapping patterns of atrophy [244]. In another mutation, Arctic APP, cortical Aβ-PET retention is very low relative to other types of ADAD mutations (PSEN1 and other APP) [249]. This mutation, however, does show abnormal glucose metabolism and CSF Aβ, t-tau, and p-tau levels. This finding supports the notion that fibrillar Aβ, as opposed to other forms of Aβ, may not be the sole harbinger for clinical Alzheimer’s disease. Cortical thinning appears in presymptomatic mutation carriers, highlighting the fast-moving progression of disease in ADAD groups. One study reported the earliest changes in the entorhinal cortex, parahippocampal gyrus, posterior cingulate cortex, and precuneus [250]. Flortaucipir-PET signal initially appears in the temporal lobe, but in contrast to sporadic cases, regional patterns of signal are also apparent in the precuneus, lateral parietal regions, and in subcortical regions of the basal ganglia [251]. Indeed, greater cortical involvement and overall higher amount of tau-PET signal in mutation carriers have been shown relative to older sporadic AD dementia cases, despite an equivalent level of cognitive impairment [251]. In a recent paper by Smith and colleagues [234], one PSEN1 mutation carrier with two Flortaucipir-PET scans prior to death in their 40 s exhibited an annual increase of approximately 20–40% signal in many cortical regions and a strong correlation with neuropathological tau markers. With regard to functional connectivity, clinically normal mutation carriers from the Colombian Kindred show right hippocampal and parahippocampal hyperactivation and less deactivation of precuneus and posterior cingulate [252], with greater activation of the anterior hippocampus reported when completing association memory functional MRI tasks [253]. As such, hyperactivity and breakdown of intrinsic functional networks in healthy mutation carriers seem to occur years prior to estimated symptom onset.

Future Research Priorities

This review has covered recent advances in Aβ, tau, and structural/function neuroimaging markers of AD and their associations with clinical and fluid markers, neuropathological staging, and models of propagation and sequencing. Even with the advent of blood-based biomarkers, AD neuroimaging will remain a critical area of research and translational evidence for the clinicopathological trajectory of AD. The techniques presented in this review can provide novel and unique findings inaccessible to other modes of inquiry and generate novel avenues of exploration in conjunction with fluid and genetic markers. It is also important to note the vast areas of neuroimaging research that have also gained traction in recent years in other critical areas of AD research, for instance PET markers of inflammation (such as ¹¹C-PK-11195 [254]) and synaptic density (such as SV2A-PET [255]), and neuroimaging of cerebral microinfarcts and vasculature [256, 257]. Much of the focus in the coming years will be on big data approaches to neuroimaging in AD [258], neuroimaging biobanks [259], and longitudinal modelling of multimodal imaging [135]. One blind spot that persists in the field is the lack of acknowledgement of certain elements of the AD pathophysiological cascade that are inherently tied to the disease but, as yet, do not have robust neuroimaging markers. For instance, major developments still need to be made in the areas of neuroimaging markers for TDP-43 [260], neuroinflammation [261], and α-synuclein [262]. There is also greater need for individualized risk profiling and precision medicine in AD, which neuroimaging can help inform [263]. In addition, major caveats to the body of work presented so far remain in relation to the racial, ethnic, and cultural diversity of the populations that are investigated, the nature and duration of longitudinal data available, and the improvement and refinement of modelling approaches and neuroimaging markers available. Another important acknowledgement in the field is the deep need for replication and validation of findings, particularly from the areas of machine learning and artificial intelligence. As blood markers take the lead as a promising new avenue for cheap and reliable biomarker detection in large swathes of the population, the advantage of neuroimaging will still remain as the stalwart for in vivo regional detection of pathology for modelling disease progression. This review has touched upon a small segment of promising advances in the field, but it leaves the reader with great anticipation for the next big advances that will aid in further understanding of the natural history of progression of the disease.

Supplementary Information

Below is the link to the electronic supplementary material.

Funding

Sole author is funded by the National Institutes of Health-National Institute on Aging Pathway to Independence Award (K99AG061238) and the Alzheimer’s Association Research Fellowship.