Applications of amyloid, tau, and neuroinflammation PET imaging to Alzheimer's disease and mild cognitive impairment

Abstract Alzheimer's disease (AD) is a devastating and progressive neurodegenerative disease for which there is no cure. Mild cognitive impairment (MCI) is considered a prodromal stage of the disease. Molecular imaging with positron emission tomography (PET) allows for the in vivo visualisation and tracking of pathophysiological changes in AD and MCI. PET is a very promising methodology for differential diagnosis and novel targets of PET imaging might also serve as biomarkers for disease‐modifying therapeutic interventions. This review provides an overview of the current status and applications of in vivo molecular imaging of AD pathology, specifically amyloid, tau, and microglial activation. PET imaging studies were included and evaluated as potential biomarkers and for monitoring disease progression. Although the majority of radiotracers showed the ability to discriminate AD and MCI patients from healthy controls, they had various limitations that prevent the recommendation of a single technique or tracer as an optimal biomarker. Newer research examining amyloid, tau, and microglial PET imaging in combination suggest an alternative approach in studying the disease process.


| INTRODUCTION
Alzheimer's disease (AD) is the most common cause of dementia worldwide. It is estimated that by 2050, 1 in 85 people worldwide will develop AD (Brookmeyer, Johnson, Ziegler-Graham, & Arrighi, 2007).
The prodromal phase of AD, defined as mild cognitive impairment (MCI), is characterised by declines in performance of one or more cognitive domains with the preservation of functional independence (Petersen, 2004). Central nervous system (CNS) degeneration and disease neuropathology predates AD and MCI. This is particularly true in presymptomatic carriers of apolipoprotein E (APOE) ε4 (Reiman et al., 2009), which is the leading genetic risk factor for AD .
The hallmark neuropathological substrates for AD and MCI are β-amyloid (Aβ) plaques and intracellular tau neurofibrillary tangles (NFTs). One major theory, the 'amyloid cascade hypothesis', suggests that the overproduction combined with dysfunctional clearance of Aβ is the fundamental event that initiates AD pathogenesis (Hardy & Selkoe, 2002). However, several lines of evidence challenge this assumption. For example, approximately 30% of healthy elderly individuals have significant levels of Aβ deposition without apparent clinical symptoms (Rowe et al., 2010). In AD, histopathological evidence suggests that Aβ levels are a poor predictor of severity of cognitive impairment (Giannakopoulos et al., 2003) and anti-amyloid interventions have demonstrated limited efficacy in clinical trials (Karran & Hardy, 2014). Presence and extent of hyperphosphorylated tau-based NFT pathology is positively associated with disease duration and severity of cognitive symptoms (Gómez-Isla et al., 1997). In addition to the more traditional markers of AD pathology, the existence of neuroinflammation in AD is currently well established. Whilst the initial inflammatory response aims to ameliorate neuronal injury, abnormally prolonged microglial activation can have detrimental effects and potentially serve to exacerbate neurodegeneration (Fakhoury, 2018).
Positron emission tomography (PET) is a neuroimaging tool designed to measure in vivo molecular processes in the brain. PET radioligands bind a target, such as a receptor, a transporter, or an enzyme. Degree of tracer binding or uptake is used to quantify neuropathology. This technology may be particularly useful for diagnostic purposes, treatment planning, and to assess disease progression in neurological illnesses (Politis & Piccini, 2012). PET biomarkers that have been recommended to improve diagnostic accuracy for AD and MCI include decreased cerebral metabolism on [ 18 F]fludeoxyglucose (FDG) PET and increased Aβ deposition on amyloid PET (Albert et al., 2011;McKhann et al., 2011). Rates of cerebral metabolism do not explicitly elucidate potential disease-causing neuropathology, but more likely highlight the degree of neuronal activity (Marcus, Mena, & Subramaniam, 2014). Characteristic region-based patterns of brain hypometabolism have been established within a range of different dementia aetiologies (Kato, Inui, Nakamura, & Ito, 2016;Minoshima, Frey, Koeppe, Foster, & Kuhl, 1995;Moonga et al., 2017;Mosconi et al., 2008;Nestor, Caine, Fryer, Clarke, & Hodges, 2003). PET tracers that measure Aβ burden, tau aggregation, and neuroinflammation provide remarkable insight directly into the processes underlying the pathophysiology of AD and MCI; however, the latter two have not yet been suggested for use in clinical practice. This review describes the recent developments and current applications of PET imaging of amyloid, tau, and neuroinflammation in AD and MCI.
Abnormalities on magnetic resonance imaging (MRI) have also been used as clinically relevant imaging markers for AD and MCI (Chandra, Dervenoulas, Politis, & Alzheimer's Disease Neuroimaging Initiative, 2018); however, this is not within the scope of the current work.
2 | AMYLOID STUDIES 2.1 | 11 C-labelled amyloid tracers Klunk et al. discovered what would become the most heavily researched amyloid PET radiotracer, 11 C-labelled Pittsburgh compound B ([ 11 C]PiB; Table 1) (Klunk et al., 2004). In their seminal study, it was shown that this benzothiazole-based radioligand was capable of discriminating between patients with a diagnosis of extremely mild AD and healthy controls.
Areas such as frontal and temporoparietal cortex, which are particularly susceptible to Aβ pathology, demonstrated increased [ 11 C]PiB retention in AD. These findings have been replicated, and moreover, recent findings highlight intermediate regional binding of [ 11 C]PiB in MCI when compared to healthy controls and AD patients (Rowe et al., 2010;Villemagne et al., 2013). PiB PET has demonstrated the ability to bind to cored and neuritic Aβ plaques, and also diffuse ones (Ikonomovic et al., 2008).
Amyloid load as measured by [ 11 C]PiB in vivo has also been associated with other markers of AD including brain atrophy (Archer et al., 2006), medial temporal hypometabolism on [ 18 F]FDG PET, impaired memory performance (Frings, Spehl, Weber, Hüll, & Meyer, 2013), and Aβ pathology from post-mortem brain tissue (Driscoll et al., 2012). Regarding disease progression, cortical [ 11 C]PiB uptake in frontal, temporal, and cingulate areas was found to be predictive of phenoconversion from MCI to AD (Brück et al., 2013). It is important to concede that MCI patients display a bimodal distribution for amyloid uptake where amyloid load is significantly elevated in a proportion of some MCI patients but not others (Hatashita et al., 2014;Nordberg et al., 2013) (Table 2). Because of this, mean levels of amyloid accumulation may not be the most reliable discriminator in MCI. In line, MCI patients with elevated binding of [ 11 C]PiB were more likely to T A B L E 1 Studies examining in vivo regional brain uptake using amyloid tracers in AD and MCI Cortical uptake in all regions including temporal, parietal, frontal, occipital, and both posterior and anterior cingulate cortex was higher in AD patients when compared to healthy controls. The posterior cingulate cortex had the best ability to discriminate between the two groups. Regional uptake values had a sensitivity of 85% and a specificity of 91% and were inversely related to global cognitive and memory performance. (Continues) convert to AD than prodromal patients who were classified as 'amyloid negative' (Okello, Koivunen, et al., 2009).
Noted strengths of [ 11 C]PiB include high amyloid selectivity and affinity, but technical limitations hinder its clinical use in centres without in-house cyclotrons such as its short decay half-life (Yeo, Waddell, Khan, & Pal, 2015). Additionally, there are significant operational costs associated with the construction and maintenance of a cyclotron on-site (Chuck et al., 2005). Data also question the sensitivity of [  T A B L E 2 Studies examining in vivo regional brain uptake using amyloid tracers in studies including amyloid positive MCI  Wolk et al., 2012). In a group of individuals with a life expectancy of no more than 6 months, [ 18 F]florbetapir showed a sensitivity and specificity of 92 and 100%, respectively, in detecting a significant degree of amyloid plaques. Moreover, the accumulation of amyloid detected by in vivo PET imaging correlated with amyloid pathology measured from brain tissue post-mortem . In AD, frontal, temporal, occipital, parietal, cingulate, and precuneus cortical areas all show increased amyloid retention as measured by this ligand compared to healthy controls. This finding also holds for MCI patients in the posterior cingulate cortex (Camus et al., 2012). A strong discriminative ability of [ 18 F]florbetapir for AD and MCI compared to a cognitively normal status is supported by a number of additional studies (Degenhardt et al., 2016;Johnson et al., 2013;Namiki et al., 2015). There is some disagreement between studies regarding the measurement of Aβ deposition using in vivo PET. This is due to a range of factors including variability in scanning time, methodology employed during analyses, identified reference regions and regions of interest, attenuation correction, partial volume correction, machines used to scan, and tracer-specific properties. This has led to some uncertainty regarding a consistent definition of abnormal amyloid levels characteristic of AD (Klunk et al., 2015). In support, Villeneuve et al. found that routinely established thresholds for Aβ positivity using [ 11 C]PiB were insensitive and likely to lead to false negatives (Villeneuve et al., 2015). There has been an attempt to solve this problem through the creation of a standardised outcome measure across amyloid-PET modalities measured in units termed 'Centiloids' ranging from 0 to 100 (Klunk et al., 2015). However, there is still a great deal of variability present in the amyloid positivity thresholds derived from this approach (Su et al., 2018 PET was unable to accurate predict the target Aβ stages in 27.84% of cases in an end-of-life cohort (Thal et al., 2018). Additionally, an amyloid positivity threshold of a standardised uptake value ratio of 1.5 was not sensitive to MCI patients in Thal Phase 1 or 2, and just lowering this value by 0.2 led to approximately 75% of healthy control subjects being classified as having significant brain amyloidosis (Ismail et al., 2019).

| TAU STUDIES
Advances in the measurement of tau pathology in humans using in vivo PET radioligands are scarce, but recent, when compared to their amyloid-specific counterparts. For years, the gold standard in quantifying tau-derived pathology was either histopathological analysis from post-mortem tissue or the invasive collection of tau and phosphorylated tau from the CSF by means of lumbar puncture (Saint-Aubert et al., 2017). This delay in applying a tau ligand was due to several major challenges that were inherently present in the imaging of tau.
Unlike amyloid, tau is intracellular, so potential ligands need to be able to cross both the blood-brain barrier and the plasma cell membrane of the neurone. In comparison to amyloid pathology, tau aggregates are present in lower concentrations throughout the brain; therefore, ligands are required to have higher specificity for tau. Additionally, tau has multiple protein conformations and isoforms, which could potentially adversely impact sites for ligand binding (Bischof, Endepols, van Eimeren, & Drzezga, 2017).
The first breakthrough in tau imaging using PET was made using the radioligand [ 18 F]FDDNP. However, this methodology concomitantly assesses amyloid plaque burden and cannot be considered a truly selective for tau pathology (Shoghi-Jadid et al., 2002), and was later shown to have poor binding affinity for NFTs (Thompson et al., 2009 This included the hippocampus (Maruyama et al., 2013). In vivo tau burden as detected by this ligand is associated with grey matter atrophy and cognitive impairment (Shimada et al., 2017).
Research suggests that radiotracers from the THK family bind to a different site than that of [ 11 C]PBB3 . The latter may be more selective to tau aggregations that have a spatial relationship with amyloid deposits, while the former may show preference for tau closely linked to brain atrophy, CSF tau, and neuropsychological functioning (Chiotis et al., 2018). The first generation of THK radio- T A B L E 3 Studies examining in vivo regional brain uptake using tau tracers in AD and MCI Notable differences in tracer uptake were observed in neocortical areas, and particularly the medial temporal cortex for those on the spectrum of AD compared to healthy controls. Medial temporal atrophy on MRI was also observed for this group. Moreover, for those along the spectrum of AD, uptake in frontal and temporo-parietal junctions were negatively associated with cognitive status, uptake in limbic, paralimbic, and frontoparietal areas were positively associated with dementia status, and uptake in frontal regions was positively associated with frontal executive dysfunction.
Chiotis et al.  (Pontecorvo et al., 2017). Increased regional tau uptake has also shown associations with impaired domain-specific neuropsychological performance including memory, language, and visuospatial abilities in numerous variants of AD (Ossenkoppele et al., 2016). Interestingly, patients with a high degree of retention in the entorhinal cortex displayed particularly poor memory functioning (Whitwell et al., 2018). A recent study also links tau burden quantified by [ 18 F]AV-1451 to neurodegeneration, specifically longitudinal brain atrophy (Das et al., 2018). Other neurodegenerative pathology shows unique profiles of tau aggregation as detected by this tracer. For example, patients diagnosed with corticobasal syndrome show retention increases in frontal and parietal cortices compared to those with MCI due to AD (Niccolini et al., 2018).

| Second-generation tau tracers
There is a fast pace of development in the tau PET imaging literature with several novel second-generation tracers currently in development and undergoing validation (Bischof et al., 2017
Moreover, the results of a genome-wide association study has impli- However, its exact role and functional significance in relation to the brain immune response is not fully understood. Under normal physiological conditions, there is a low expression of TSPO limited to glial cells. Nevertheless, during neuronal injury or insult, when microglia are activated, TSPO levels in turn experience a significant upregulation (Rupprecht et al., 2010).
Moreover, immunohistochemistry studies have confirmed that TSPO upregulation and microglia activation co-localise spatially following a neurotoxic intervention, suggesting that TSPO can measure neuroinflammation through the detection of activated microglia (Kuhlmann & Guilarte, 2000). Beyond microglia, increased TSPO expression is also observed in reactive astrocytes (Rupprecht et al., 2010). While numerous TSPO radioligands have been developed so far, we will focus on those investigated in AD/MCI patient populations.

| [ 11 C]PK11195
The most thoroughly researched TSPO radiotracer is [ 11 C]PK11195 (Table 4). In a pioneering study, Cagnin et al. demonstrated that patients with AD had a signature pattern of [ 11 C]PK11195 uptake in brain areas that included the cingulate, temporoparietal, and entorhinal cortex (Cagnin et al., 2001). Since then, results from this tracer have been mixed and somewhat contradictory. Minimal or small clusters of increased binding in MCI and mild to moderate AD was reported in two studies (Schuitemaker et al., 2013;Wiley et al., 2009).
These authors speculated that either microglial activation is implicated later in the disease course or [ 11 C]PK11195 is not a sensitive marker of such activity. However, these claims are challenged by findings from studies that used region of interest-based analyses to show widespread tracer uptake throughout the cortex for AD patients (Edison et al., 2008;Fan, Aman, et al., 2015;Passamonti et al., 2018), particularly in parietotemporal areas (Yokokura et al., 2011) and for MCI patients especially in the frontal cortex (Fan, Aman, et al., 2015;. The neurodegenerative effects of this potentially prolonged immune response are hinted at when considering that increased tracer retention has been linked with hippocampal atrophy (Femminella et al., 2016). Utilising [ 11 C]PK11195, longitudinal research suggests an initial reduction of microglial activation in prodromal disease stages (Fan et al., 2017) contrasting with a subsequent increase in microglial activation in AD during disease progression (Fan, Okello, et al., 2015).

Results pertaining to the relationship between [ 11 C]PK11195
binding and cognitive status are also conflicting. Three studies found inverse correlations between MMSE score and tracer uptake (Edison et al., 2008;Fan, Aman, et al., 2015;Yokokura et al., 2011) and one demonstrates a negative correlation between tracer uptake in the precuneus and episodic memory performance measured by the Rey Auditory Verbal Learning Test (Passamonti et al., 2018), which suggests a role for microglial activation in disease severity. However, others failed to find similar associations between [ 11 C]PK11195 retention and neuropsychological performance (Schuitemaker et al., 2013;Yokokura et al., 2017). This pattern of contradictory results could possibly be attributed to small sample sizes, limitations of the tracer and variability in study methodology.
Several technical limitations of [ 11 C]PK11195 were found, including its short half-life of approximately 20 minutes that is a barrier for centres without a costly on-site cyclotron, low brain uptake, and importantly low signal to noise ratio. It also exhibits a high degree of nonspecific binding (Ching et al., 2012). This nonspecific binding may apply to targets including brain-based lipids (Hatty et al., 2014) and α1-acid glycoprotein (Lockhart et al., 2003). Additional clinical translation difficulties are imposed on the molecule through carbon-11 labelling (Chauveau, Boutin, Van Camp, Dollé, & Tavitian, 2008). However, suboptimal modelling of this tracer is likely its most pressing concern as there has been substantial difficulty in the definition of a true reference region, which is an area absent of binding, for [ 11 C]PK11195 (Chauveau et al., 2008). When using an arterial plasma input function, the use of a reference region allows for valid quantification of binding potential (Cunningham, Parker, Rabiner, Gee, & Gunn, 2005). In fact, variability in the findings of studies discussed using [ 11 C]PK11195 could, at least in part, reflect the use of various and inconsistent methods for reference region quantification. To account for this issue, an automatic supervised T A B L E 4 Studies examining in vivo regional brain uptake using TSPO tracers in AD and MCI No difference in TSPO binding was found when comparing groups on the AD spectrum with controls in any brain region. Edison et al. (2008) [ 11 C]PK11195 13 AD patients, 10 healthy controls Relative to healthy controls, areas in frontal temporal, parietal, and occipital association cortex, in addition to the cingulate and striatum, showed increased tracer uptake in AD patients. Inverse correlations between uptake in posterior cingulate, parietal, and frontal cortical areas and global cognition were found. Passamonti et al. (2018) [ 11 C]PK11195 16 AD and MCI patients, 13 healthy controls In a combined group of AD and MCI patients, increased binding was found in brain areas within the occipital, parietal, and temporal cortex, in addition to medial temporal regions including the hippocampus and amygdala. Binding in the precuneus was negatively associated with performance on a measure of delayed recall. Regions including the temporal, frontal, orbital, straight, parietal gyrus, insula, putamen, and occipital lobe were 28-36% higher in MCI relative to controls. Throughout the four lobes of the cortex, and the insula, thalamus, and hippocampus microglial activation and amyloid load measured by [ 11 C]PiB were positively correlated, a similar but not as widespread relationship was evidenced in MCI. TSPO uptake in frontal, temporoparietal, and occipital cortex was negatively correlated with global cognition, while regional associations were also found with cerebral glucose hypometabolism on FDG PET. Patients with AD, including the prodromal form, demonstrated greater tracer uptake in regions that include the precuneus, parietal, temporal cortex, and medium and posterior cingulate compared to controls. This uptake was positively associated with performance on global cognition and grey matter volume. This also holds for regional amyloid uptake. Suridjan et al. (2015) [ 18 F]FEPPA 21 AD patients, 21 healthy controls Temporal, frontal, parietal, and occipital cortical regions and the hippocampus demonstrated increased tracer retention for AD patients compared to controls. This also held for the posterior limb of the internal capsule and the cingulum bundle. Regional uptake was associated with impairment in visuospatial ability and language ability.

Yokokura
clustering procedure utilising a priori kinetic classes has been developed to extract grey matter estimates that can be reliably classified as reference region tissue (Turkheimer et al., 2007).

| Second-generation TSPO tracers
The shortcomings of [ 11 C]PK11195 were overcome with the advent of second-generation TSPO ligands, which notably had an improved signal-to-noise ratio and higher binding affinity compared to [ 11 C] PK11195 (Edison & Brooks, 2018). While many second-generation TSPO tracers have been discovered, we will cover only those that have been most widely used in humans and specifically in AD or MCI.  (Hamelin et al., 2016;Kreisl et al., 2013;Lyoo et al., 2015;Suridjan et al., 2015;Varrone et al., 2015;Yasuno et al., 2008). The temporal pattern of neuroinflammation over the course of the AD has also been well-characterised through longitudinal investigations. In AD patients, yearly average increases in TSPO binding that ranged from 2.5 to 7.7% were shown using [ 11 C]PBR28 (Kreisl et al., 2016), whilst for [ 18 F]DPA-714 an elevated annual change of 13.2% in tracer binding was displayed (Hamelin et al., 2018).
A change in role of activated microglia is supported by a large longitudinal study utilising [ 18 F]DPA-714 (Hamelin et al., 2016;Hamelin et al., 2018). Participants with MCI and higher initial TSPO binding had a slower rate of decline measured by the Clinical Dementia Rating and smaller increase in TSPO binding than those with lower initial TSPO binding. These results, coupled with the aforementioned [ 11 C] PK11195 studies, have led to the proposal of a dual peak hypothesis of neuroinflammation in AD (Fan et al., 2017). This suggests that the early peak in activated microglia in MCI patients is initially protective, attempting to remove Aβ, whereas the later peak in activated microglia is detrimental. Associations between TSPO expression and clinical outcome for individuals on the spectrum of AD may only be observable for neuroimaging data collected during rising rather than declining phases of these peaks. It is important to note that results from TSPO imaging studies that include MCI patients may reflect either a rising or declining PET signal. While different phenotypes of activated microglia are detectable in pathological studies (Tang & Le, 2016), PET imaging utilising TSPO is unable to differentiate the microglial subtype.
Similar to [ 11 C]PK11195, in vivo increases in TSPO binding are associated with impairments in global cognition and memory (Hamelin et al., 2018;Kreisl et al., 2013), but also extend to domains that include visuospatial and language ability, and executive functioning, in addition to dementia severity and brain atrophy (Hamelin et al., 2018;Kreisl et al., 2013;Kreisl et al., 2016;Suridjan et al., 2015). However, some studies demonstrate no such correlations with the ADAS-Cog or MMSE (Yasuno et al., 2008;Yasuno et al., 2012)  upregulated cortical tracer retention when compared to healthy controls, especially in the temporal lobe Yasuno et al., 2012), and others showing no such significant differences (Knezevic et al., 2017;Kreisl et al., 2013).
It is imperative to acknowledge the primary limitation of these second-generation tracers, which is their particular sensitivity to a single nucleotide polymorphism (SNP) of the TSPO gene. Genetic variation in the SNP rs6917 results in different patterns of binding affinity to TSPO: high affinity binders, low affinity binders, and mixed affinity binders. Therefore, it is essential to test for these genetic polymorphisms when examining data from these tracers and exclude low affinity binders, which consist of approximately 10% of the population, from analyses (Edison & Brooks, 2018;Yoder et al., 2013). Not being influenced by this TSPO polymorphism can be considered an advantage of [ 11 C]PK11195.

| INTERPLAYS BETWEEN TAU, AMYLOID, AND MICROGLIA
Overall, tau, amyloid, and TSPO radiotracers show good ability in the detection of AD and MCI, in addition to associated neuropathology (Table 5) Hamelin et al., 2016;Knezevic et al., 2017;Parbo et al., 2017;Parbo et al., 2018). Whilst older and using smaller samples, studies indicating negative or no association between microglial activation and Aβ accumulation (Kreisl et al., 2013;Wiley et al., 2009;Yokokura et al., 2011) highlights the need for future research to clarify the nature of this relationship using in vivo brain imaging technology.
Two very recent studies have sought to investigate whether associations also exist between TSPO binding and both tau retention on [ 18 F]AV-1451 and amyloid load. The first found no association between the two, but used a sample of patients primarily with prodromal disease (Parbo et al., 2018). The second study evidenced a positive relationship between neuroinflammation and both tau and amyloid pathology in patients with AD and MCI, with similar targeted clusters of cortical regions . Furthermore, the correlation between tau and neuroinflammation was demonstrated even in participants without significant amyloid burden, suggesting a process independent of Aβ. These results offer the first in vivo evidence in AD and MCI patients that neuroinflammation and tau pathology have a pathophysiological link.

| CONCLUSIONS
Molecular imaging with PET has shed light into the complex interplay between Aβ, tau, and neuroinflammation in AD and MCI and helped to clarify to what extent these are part of the normal ageing process or if they represent a distinct pathophysiological process. It has also assessed the patterns of neuropathological regional depositions and their relation to cognitive decline, disease progression and, ultimately, neurodegeneration.
In vivo imaging of dementia-related pathology through the reviewed PET radioligands has several benefits. First, a preclinical detection of the disease can be achieved by examining early molecular changes. In addition, the differential diagnosis process between AD and other neurodegenerative disorders can be refined through a focus on characteristic neuropathological mechanisms. From a research perspective, these techniques allow for the study of relationships T A B L E 5 Studies examining sensitivity and specificity (in %) of amyloid, tau, and TSPO tracers in the detection of AD and neuropathology   between cognition and specific neurodegenerative processes across the AD continuum. Finally, these PET molecular targets can serve as