Background

Alzheimer’s disease (AD) is the leading cause of dementia worldwide and it is pathologically characterised by the deposition of extracellular β-amyloid plaques and intracellular neurofibrillary tangles of hyperphosphorylated tau proteins [1]. Pathological changes may start up to 20 years before the onset of symptoms [2,3,4,5]. To date, treatments have only been approved for the dementia stage of the disease and provide modest symptomatic benefit but no slowing of progression. A global research priority is to find therapies to prevent or delay symptom onset and functional decline [6], and a number of candidate agents have shown promise in cell or animal models. Despite this, clinical trials of potential disease-modifying treatments have proven unsuccessful thus far. This may be partly due to limitations of the treatments (e.g. incorrect molecular target or inadequate target engagement, dose, or duration). However, the failures could also reflect inappropriate trial populations. Typically, trials have recruited patients with mild to moderate dementia in whom the disease process may be too advanced to be amenable to treatment. An alternative approach is to treat subjects at risk of AD dementia, which is the aim of secondary prevention—preventing neurodegeneration, cognitive decline, and dementia [7, 8].

Research criteria from the International Working Group (IWG) [9, 10] and the National Institute on Ageing-Alzheimer Association (NIA-AA) [11,12,13] propose the use of biomarkers to define pre-clinical AD as the disease stage characterised by amyloid pathology, with or without neurofibrillary tangles and/or features of neurodegeneration, even in the absence of clinical manifestations. Mild cognitive impairment (MCI) is used to denote an intermediate stage between normal cognition and dementia in which subjects have objective cognitive impairment in the absence of functional disability interfering with daily activities. MCI with biomarker evidence of AD has been termed prodromal AD under the IWG criteria [9, 10] and MCI due to AD by the NIA-AA in 2011 [12], but the 2018 research framework does not take into account syndromic diagnosis [11]. Non-demented subjects with evidence of amyloid pathology, i.e. subjects with AD without dementia [14], provide an opportunity for intervention prior to irreversible neuronal loss. Designing trials for subjects in pre-dementia stages of AD is greatly facilitated by the ability to identify subjects at increased risk of cognitive decline and progression to dementia. In the earliest stage, neuropsychological testing to detect subtle cognitive abnormalities in the absence of obvious symptoms has some predictive value [15], but biomarkers that are sensitive to underlying pathological change could further increase prognostic accuracy. Ideally, early-stage biomarkers should predict risk and likely timing of cognitive decline and progression to dementia in a reliable, non-invasive, and cost-effective manner. Secondly, trials in non-demented subjects would benefit from biomarkers that can monitor treatment effects and should thus reflect disease progression, be sensitive to detect pathologically significant changes over time and in response to treatment, and be highly reproducible and reliable in a multi-centre setting. Neuroimaging techniques have developed rapidly over the past decade and they currently offer a comprehensive armamentarium that can be employed to address this unmet need. Over and above structural neuroimaging markers that are widely available and used to support AD diagnosis at the dementia stage, advances in imaging techniques allow the detection and quantification of molecular, functional, and structural brain changes that precede gross atrophy.

In this paper, we discuss the use of neuroimaging markers in subject selection for inclusion or stratification in secondary prevention trials, their potential to serve as outcome markers in trials, and for monitoring trial safety. The overall aim is to devise a neuroimaging strategy that will maximise the information required to enrol and monitor secondary prevention trials in AD (including progression from MCI to dementia). To identify potential imaging markers, a comprehensive review of the existing literature was performed. Based on the evidence from the literature and the personal experience of the authors, recommendations are centred around three domains: 1) imaging markers for subject selection and stratification; 2) imaging markers with potential to be used as outcome measures in clinical trials; and 3) imaging markers for monitoring trial safety. The utility of longitudinal imaging as a run-in for clinical trials, identification of exclusion criteria, and vascular comorbidity were also taken into account.

Methods

Neuroimaging modalities: from molecular changes to structural damage

The past two decades have seen major advances in neuroimaging. Different imaging modalities are now able to track different aspects of the hypothesised pathological cascade of events in AD in vivo [3] (Box 1, Figs. 1, and 2). Positron emission tomography (PET) with amyloid-specific tracers, and more recently also tau-binding ligands, can visualise and quantify molecular pathology at an early stage. Alterations in functional imaging biomarkers, reflecting early synaptic dysfunction and neuronal injury, can be measured with various PET and magnetic resonance imaging (MRI) measures. Collectively, molecular and functional changes may lead to synaptic loss, inflammation, white matter damage, and neuronal cell death, eventually leading to macroscopic changes such as regional and global brain atrophy seen on structural MRI. Increasing evidence demonstrates that cerebrovascular changes have an additive effect on neurodegeneration, accelerate cognitive decline and progression to dementia, and may even be part of the pathological cascade of AD [16,17,18]; hence, we also address MRI markers of vascular pathology.

Fig. 1
figure 1

PET imaging biomarkers. Examples of normal (top) and abnormal (bottom) positron emission tomography (PET) imaging markers in three different subjects. For all images, the warmer the colour, the more tracer binding. Left: amyloid PET with [18F]-flutemetamol. In the abnormal scan, diffuse tracer binding to fibrillary amyloid can be observed. Middle: tau PET with [18F]-AV-1451. In the abnormal scan, tracer binding to tau can be observed in the temporal lobes. Right: Fluorodeoxyglucose (FDG)-PET scan. In the abnormal scan, there is hypometabolism of the parietal lobes

Full size image
Fig. 2
figure 2

MRI imaging biomarkers. Left: T1-weighted MRI (top) showing severe hippocampal atrophy and example of diffusion tensor imaging (DTI) (bottom). Middle: example of functional imaging markers with arterial spin labelling (ASL) (top) and resting state functional magnetic resonance imaging (rs-fMRI) (bottom). Right: imaging of vascular pathology with thalamus lacune on T2 (top; arrow) and white matter hyper-intensities on fluid attenuated inversion recovery (FLAIR) (bottom)

Full size image

Search strategy and selection criteria

References for this review were identified by searching the PubMed/Medline database in August 2017. Relevant articles were identified using the following search terms alone and in varying combinations: “amyloid PET”, “tau PET”, “MRI”, “structural MRI”, “functional MRI”, “FDG PET”, “fluorodeoxyglucose PET”, “TSPO PET”, “diffusion tensor imaging”, “arterial spin labeling”, “magnetic resonance spectroscopy”, “cognitively healthy”, “normal cognition”, “mild cognitive impairment”, “subjective cognitive decline”, “Alzheimer’s disease”. Papers published in English were included. Further references were obtained by screening references from retrieved articles and on the basis of the personal knowledge of the authors. In the case of topics already extensively covered in the literature, as was often the case for MCI, a reference article or review was selected by the authors. The final selection of articles was based on relevance to the topics covered in this review, as judged by the authors

Subject selection and stratification

Various planned and ongoing clinical trials for AD primarily target the amyloid cascade, aiming at the removal of amyloid plaques or prevention of misfolding of amyloid into the β conformation [19]. We recommend that trials targeting the amyloid pathway should include methods to recruit subjects with evidence of an appropriate level of amyloid pathology. Various PET tracers are capable of measuring and spatially localising β-amyloid deposits. Alternatively, β-amyloid peptides may be measured in cerebrospinal fluid (CSF). Studies have demonstrated good concordance between CSF β-amyloid 1–42 and amyloid PET measures, even though these markers might represent different pools of amyloid in the brain [20]. Several studies suggest that CSF β-amyloid levels become abnormal prior to an amyloid PET signal [21,22,1: Table S1). Imaging changes associated with other neurodegenerative disorders are, however, unlikely to be present in the early stages, which may preclude the reliable exclusion of these subjects.

Fig. 3
figure 3

Step-wise approach for subject inclusion and testing. Information from clinical measurements (and, in the near future, possibly also plasma) may be used to select subjects with an increased risk of amyloid pathology (screening). Provided there are no exclusion criteria, molecular measurements of amyloid (or tau, depending on the treatment target) can be used to screen-in subjects for clinical trials. Finally, imaging measures predicting imminent cognitive decline may be used for additionally enrichment. APOE apolipoprotein E, CSF cerebrospinal fluid, MRI magnetic resonance imaging, PET positron emission tomography

Full size image

Structural and functional imaging markers can also be used to identify subjects at risk of imminent cognitive decline, which will be reviewed in the following paragraphs. This is especially relevant for phase 3 trials in subjects with pre-clinical AD targeting cognition as a primary outcome. Enrichment of clinical trials targeting clinical end-points by means of amyloid PET and/or MRI may reduce sample sizes and costs, as shown in subjects with MCI [27, 28], but this work needs to be extended to the pre-clinical phase. Table 1 summarises the available evidence for the use of different imaging markers for subject selection in clinical trials.

Table 1 Summary of evidence for use of imaging markers for subject selection and as outcome measures in clinical trials in pre-dementia Alzheimer’s disease
Full size table

Molecular imaging

Amyloid PET: predictor of decline?

Amyloid pathology measured with PET is an established prognostic marker in subjects with MCI (sensitivity 82% (95% confidence interval (CI) 74–88) and specificity 56% (95% CI 49–64) to distinguish stable MCI patients from those who progress to dementia) [29]. In cognitively normal subjects, amyloid positivity has been associated with an increased risk of cognitive decline and progression to dementia in several longitudinal studies [30,31,32,33,34,35,36,37,38,39,40,41], although studies with sufficiently long follow-up and large sets of data to establish the exact risk are required [40,41,42,43] (Table 2). The method of choice to classify subjects as amyloid-positive or amyloid-negative remains a matter of debate (Box 2). Recent evidence has also suggested that amyloid plaques might follow consistent deposition patterns in different regions of the brain, making it possible to stage amyloid pathology [44]. Although the relationship between amyloid positivity and later cognitive decline in cognitively normal subjects has been established, it has been suggested that the rate at which this occurs depends on the presence of neurodegeneration [45, 46]. Amyloid positivity is also consistently associated with increased brain atrophy rates in cognitively normal subjects (Additional file 1: Table S2). Hence, amyloid pathology is a necessary factor to assess whether an individual will decline due to AD pathology but is not sufficient to stage disease, or to predict when and how fast the decline will occur, since the timing depends on the rate of neurodegeneration [41, 46,47,8]).

Structural imaging

Grey matter atrophy

Patients with AD dementia show a pattern of widespread cerebral atrophy. Measures of global cortical atrophy have been used as an outcome marker in clinical trials in MCI and AD dementia (e.g. [172,173,174]). Besides global measures, regional changes can also be examined. Regional changes in different brain regions will likely relate more or less strongly to changes in different cognitive instruments, depending on the cognitive domains they subserve [175].

Hippocampal atrophy rates are a good candidate to serve as an outcome marker in multi-centre clinical trials, as long as standardisation of image acquisition between centres, consistent within-subject scanner acquisition, and uniform post-processing methods are performed. Test-retest reliability of repeated manual and automated hippocampal measurements from the same scanner is usually high (test-retest variability 1–4%) [176, 177], but some algorithms are more robust than others [178]. Efforts to standardise the measurement of hippocampal volumetry in multi-centre studies have been undertaken and it is now standard practice for trials to employ single algorithms and centralised analysis [89, 179]. With the establishment of the Harmonised Protocol for hippocampal segmentation, there is a new gold standard against which automated measurement may be validated [180]. Variability between scanners using the same acquisition protocol and measurement algorithm is low [176]. However, the agreement in terms of absolute volumes varies with acquisition protocols and field strength. For example, change in voxel size can lead to systematic errors in the range of 5% for hippocampal volume [181]. Methods to correct for these variabilities are being investigated [86].

Using data from the Australian Imaging, Biomarker and Lifestyle (AIBL) study, a sample size of 384 subjects per arm was estimated to be needed to detect 25% slowing of hippocampal atrophy rates over 18 months in subjects with pre-clinical AD with a power of 80% [182]. These numbers will be larger when taking into account atrophy occurring with normal ageing, which should be considered [183]. Measures of hippocampal atrophy rates have already been used as (secondary) end-points in various clinical trials in MCI and mild AD dementia, with varying results [151].

Diffusion tensor imaging

To date, there is little evidence for the use of DTI measures in (multi-centre) clinical trials. Several studies have pointed to the variability of DTI measurements in multi-centre studies and the need for extensive site harmonisation and calibration prior to starting [184,185,186]. More work is needed on the optimisation of DTI measurements in multi-centre settings, which is now being addressed [187]. So far, longitudinal DTI measures have only sparsely been used as (secondary) outcomes in clinical trials [188].

Monitoring of side-effects

Amyloid-related imaging abnormalities

MRI scans play an important role in safety monitoring during clinical trials. An issue of particular importance in the Alzheimer’s field is the occurrence of amyloid-related imaging abnormalities (ARIA) that have emerged in several anti-amyloid immunotherapy trials [189], although ARIA also occurs spontaneously—including in placebo arms [190]. ARIA consists of ARIA-E (parenchymal oedema or sulcal effusion) with signal hyper-intensities on FLAIR, and ARIA-H (cerebral microbleeds or superficial siderosis) with hypo-intensity due to hemosiderin on T2*-weighted or susceptibility-sensitive pulse sequences (Fig. 4). Adequate training for radiological reads is recommended for both ARIA-E and ARIA-H to ensure reliable detection of subtle cases and to maximise consistency between raters [191]. Severity of ARIA-E can be rated using dedicated rating scales [192]. Central assessment of the images is recommended to guarantee quality control and to minimise differences in visual inspection and quantification.

Fig. 4
figure 4

Amyloid-related imaging abnormalities. Example of ARIA-E on FLAIR with sulcal effusion (left) and ARIA-H with multiple microbleeds (middle) and superficial siderosis (right) on T2* images

Full size image

The detection of ARIA-H is dependent on the pulse sequence contrast mechanism and parameters (e.g. echo time and slice thickness) as well as field strength [189]. The reported prevalence of microbleeds in AD dementia subjects ranges from 18% at 1 T to 71% at 7 T [193]. Since the presence of cerebral microbleeds confers a risk for future haemorrhage and development of ARIA, many trials limit the number of microbleeds at entry and monitor development of new microbleeds during the trial. A key consideration in the management of emergent ARIA cases during a trial is whether any clinical symptoms are associated with the radiological observations.

In pre-clinical AD populations, the baseline incidence and the frequency of spontaneous emergent ARIA is less well characterised than in clinically demented subjects, although one study observed a lower incidence than in MCI and AD dementia populations [194]. Although several large randomised clinical trials in pre-clinical AD populations have recently commenced, the ARIA findings are not yet available to the wider community.

Regulatory perspective

In 2011, following observations in the bapineuzumab trials [195] and recommendations from an Alzheimer’s Association Round Table workgroup [189], the US Food and Drug Administration (FDA) requested the adoption of ARIA-based exclusion and discontinuation criteria for amyloid-targeting therapies. These limited the enrolment to subjects with at most four microbleeds at baseline and defined minimum acquisition standards. Since that time, for non-amyloid targeting mechanisms or if the ARIA risk for a particular compound or mechanism has been discharged in earlier trials, these criteria have been relaxed in some trials. Nevertheless, these recommendations have become a de facto standard. Systematic data collection and comparison between 2D-T2* and more sensitive sequences (e.g. 3D susceptibility-weighted imaging) is lacking. A well-controlled head-to-head comparison of the above T2* sequence prescription with a more sensitive alternative sequence would further the field’s understanding of ARIA prevalence and evolution in the natural history and provide data to help establish meaningful and comparable cut-off criteria for these alternative sequences early in the disease course.

Conclusions and practical implementation

Neuroimaging allows visualisation of many aspects of the pathological cascade of AD, including the presence of (pre-clinical) molecular pathology and downstream functional and structural markers of neurodegeneration before the onset of dementia. Information derived from imaging can aid in identifying non-demented subjects with an increased risk of future cognitive decline and disease progression to be included in secondary prevention trials. Combining information on underlying Alzheimer’s pathology from amyloid PET (or CSF) with markers of neurodegeneration from structural MRI (or FDG-PET) provides the optimal strategy to identify subjects who are at highest risk of cognitive decline within the typical time frame of a clinical trial. Advanced functional and structural imaging techniques to predict cognitive decline at a pre-dementia stage hold promise but await further research and validation. Neuroimaging can help to characterise subjects in terms of comorbidities (e.g. cerebrovascular disease) or to define more homogeneous subgroups that can be used for stratification. Neuroimaging can also serve as a (secondary) outcome marker in trials. Amyloid PET can be used to assess target engagement for pharmacological compounds targeting the amyloid cascade, and structural imaging (or FDG-PET) can be used to assess possible treatment effects on neurodegeneration. Finally, imaging is crucial for monitoring safety and potential side effects, such as ARIA, during trials.

Trials in non-demented subjects are greatly facilitated by the establishment of a cohort of well-phenotyped subjects that can be included in clinical trials. This is an approach that has been taken by the European Prevention for Alzheimer’s Dementia (EPAD; http://ep-ad.org/) study [196]. Within EPAD, subjects at elevated risk for AD are identified from various parent cohorts throughout Europe and enrolled into a longitudinal observational cohort study to serve as a trial-readiness cohort for proof-of-concept intervention studies. In addition to screening subjects, this trial-readiness cohort also provides an opportunity to apply run-in designs in which pre-trial longitudinal imaging can be used to determine within-subject rates of change, which increases the statistical power to detect treatment effects and reduces required sample sizes [197].

There are no formal guidelines on the use of neuroimaging measures in such trial-readiness cohorts of non-demented subjects. Some experience has been gathered through longitudinal imaging from multi-centre studies with large cohorts such as ADNI, with a strong focus on the MCI and dementia stages, and more recently with the AIBL study, with a focus on cognitively normal subjects [198, 199]. In ADNI, the imaging protocol initially included structural MRI and a subgroup with amyloid PET, and was later complemented by FDG-PET and advanced MR techniques such as ASL, DTI, and rs-fMRI in ADNI-2. Subjects in ADNI-3 also undergo tau PET. In AIBL, the imaging protocol includes structural MRI and amyloid PET. Based on the literature reviewed in this paper, experience gained in other studies, and practical considerations, neuroimaging recommendations for the EPAD longitudinal cohort study have been formulated (Table 4). These recommendations are based on a combination of desired information richness, patient burden, stratification capabilities, and provision of run-in data for trials. The recommended core imaging protocol for the EPAD longitudinal cohort study consists of yearly MRI scans including isotropic 3D-T1 and 3D-FLAIR sequences, as well as a short 2D-T2W and 2D-T2* sequence to assess neurodegenerative and vascular pathology. Advanced sequences such as 3D-T2*/SWI, DTI, ASL, and rs-fMRI will be acquired in subsets of patients depending on site interest and experience with acquisition methodology. Centralised assessment of the scans will be performed to guarantee quality control and maximise consistency in visual rating scales and quantification. All subjects in EPAD will undergo lumbar puncture to assess amyloid pathology in CSF. A large subgroup will also undergo amyloid PET, financed through the sister project Amyloid Imaging to Prevent AD (AMYPAD; http://amypad.eu/). Amyloid PET was prioritised as molecular imaging based on its potential to confirm and localise amyloid pathology, its broad availability and standardisation, and given that the initial molecular targets in upcoming proof-of-concept trials will likely address the amyloid pathway. Static amyloid PET imaging is currently the most common approach in clinical trials. Dynamic imaging, however, may be preferable to determine the true binding potential rather than an SUVr that may be biased by the choice of the reference region and flow effects. This argument becomes more relevant for longitudinal imaging where changes can be small and in treatment trials where flow alterations may occur. In addition, data from the initial uptake of the tracer can be used as a proxy to measure cerebral blood flow. As FDG-PET would add an additional radiation dose, it has not been included in the EPAD imaging protocol. While tau PET imaging is emerging, current implementation is limited due to costs, availability, and lack of standardisation. Development and accessibility of tau PET is encouraged as a potentially more proximate biomarker and predictor of disease progression and as therapies targeting tau enter clinical trials.

Table 4 Imaging recommendations for EPAD longitudinal cohort study
Full size table