**Preface to "Radiolabelled Molecules for Brain Imaging with PET and SPECT"**

Positron emission tomography (PET) and single-photon emission computed tomography (SPECT) are in vivo molecular imaging methods which are widely used in nuclear medicine for diagnosis and treatment follow-up of many major diseases. These methods use target-specific molecules as probes, which are labeled with radionuclides of short half-lives that are synthesized prior to the imaging studies. These probes are called radiopharmaceuticals. Their design and development is a rather interdisciplinary process covering many different disciplines of natural sciences and medicine. In addition to their diagnostic and therapeutic applications in the field of nuclear medicine, radiopharmaceuticals are powerful tools for in vivo pharmacology during the process of preclinical drug development to identify new drug targets, investigate pathophysiology, discover potential drug candidates, and evaluate the in vivo pharmacokinetics and pharmacodynamics of drugs.

The use of PET and SPECT for brain imaging is of special significance since the brain controls all the body's functions by processing information from the whole body and the outside world. It is the source of thoughts, intelligence, memory, speech, creativity, emotion, sensory functions, motion control and other important body functions. Protected by the skull and the blood–brain barrier, the brain is somehow a privileged organ with regard to nutrient supply, immune response, and accessibility for diagnostic and therapeutic measures. Invasive procedures are rather limited for the latter purposes. Therefore, noninvasive imaging with PET and SPECT has gained high importance for a great variety of brain diseases, including neurodegenerative diseases, motor dysfunctions, stroke, epilepsy, psychiatric diseases, and brain tumors. This Special Issue focuses on radiolabeled molecules that are used for these purposes, with special emphasis on neurodegenerative diseases and brain tumors.

Molecular imaging of neurodegeneration has become a useful noninvasive clinical tool to early detect pathophysiological changes in the brain and is regarded to be of special importance for prognostic purposes, therapeutic decision making, and therapy follow-up. Alzheimer's disease (AD) and Parkinson's disease (PD) are regarded as the most common and known neurodegenerative disorders, with a growing impact especially in countries with rapidly increased life expectancies during the last decades. Misfolded proteins such as β-amyloid, τ -protein, α-synuclein together with neuronal dystrophy characterize the main pathology of these diseases. Furthermore, multiple neurotransmitter systems are affected and involved in the cellular pathology.

The initial review written by Uzuegbunam, Librizzi, and Yousefi provides an overview of the currently available PET radiopharmaceuticals, examining the timeline and important moments that led to the development of these tracers and offering an outlook that is especially focused on the design of α-synuclein-targeting radiotracers.

This review is followed by a number of articles describing other potential targets for diagnostic and/or therapeutic approaches towards AD and PD. Neuronal dystrophy in AD is accompanied by a reduced glucose metabolism, which can be measured with PET using the radiopharmaceutical 2 deoxy-2-[<sup>18</sup>F]fluoroglucose ([<sup>18</sup>F]FDG). However, at the time at which significant reductions of [ <sup>18</sup>F]FDG accumulation in brain regions become evident, AD has usually progressed into the clinical stage. In order to prevent and/or start early treatment of AD, disease diagnosis during the preclinical stage is needed. To address this issue, a novel metabolic brain network analysis of FDG-PET using

kernel-based persistent features was proposed by Kuang et al. The FDG imaging data from 140 subjects with AD, 280 subjects with mild cognitive impairment, and 280 healthy normal controls suggest that the approach has the potential of an effective preclinical AD imaging biomarker.

Synaptic loss is well established as the major structural correlate of cognitive impairment in AD. The ability to measure in vivo synaptic density could accelerate the development of disease-modifying treatments for AD. The synaptic vesicle protein 2 (SV2) is involved in synaptic vesicle tracking and regarded as a potential biomarker for the measurement of synaptic density. It consists of the three isoforms, A, B, and C, whereby SV2A, in particular, has been closely related to AD. Therefore, the selectivity of a radiopharmaceutical towards these different isoforms is an important issue. The article of Serrano et al. evaluates the in vivo specificity of [ <sup>18</sup>F]UCB-H, a radiotracer with nanomolar affinity for human SVA2, by comparing the SV2A protein with SV2B and SV2C using microPET in rats.

The potential of nicotinic acetylcholine receptors (nAchRs), as indicators of cholinergic neuronal functions, has previously been reported by a variety of papers, including those of our group, to be reduced in AD and PD. In this Special Issue, the dibenzothiophene derivatives [ <sup>125</sup>I]Iodo-ASEM and [ <sup>18</sup>F]ASEM, isomers of our own ligand [ <sup>18</sup>F]DBT10 (previously published in Molecules 20, 18387-421, 2015), were preclinically characterized in pigs as suitable radiotracers for the imaging of homo- and heteromeric α7 nAchRs with PET and SPECT.

The sleep–wake cycle in patients with AD has been associated with τ pathology and the dysregulation of the neuropeptide orexin, which exerts its action by binding to orexin receptors 1 and 2. There is evidence that the OX2R gene's rs2653349 and rs2292041 polymorphisms may be associated with AD. The FDA has approved orexin as a drug to treat insomnia. For these and other reasons, imaging of the orexin receptor status with PET and/or SPECT appears to be highly impactful. Bai et al. report a new PET radiotracer for orexin receptors neuroimaging which was preclinically used for PET investigations in mice and monkeys.

The following article deals with the adenosine A2A receptor (A2AR), which is regarded as a particularly appropriate target for the non-dopaminergic treatment of PD. Schr¨oder et al. selected the known A2AR-specific radiotracer [ <sup>18</sup>F]FESCH and developed a simplified two-step one-pot radiosynthesis, in order to promote its clinical applicability. The radiotracer was used to investigate the suitability of rotenone-treated mice as an animal model of PD.

In a previous issue (Molecules 21, 650, 2016), the development of <sup>18</sup>F-labelled PET ligands for the molecular imaging of the cyclic nucleotide phosphodiesterase 2A, a key enzyme in the cellular metabolism of the second messengers cAMP and cGMP, was reviewed, and PDE2 was proposed as a viable target for future drug development for AD, PD, Huntington's chorea and psychiatric diseases. Two articles by Ritawidya et al. dealing with fluorine-containing benzoimidazotriazine-based PDE2A-selective ligands for potential PET imaging are included in this Special Issue—the first describes the synthesis and in vitro evaluation of 8-pyridinyl-substituted benzo[e]imidazo[2,1-c] [1,2,4]triazines as selective PDE2A inhibitors, and the second describes the radiosynthesis and biological evaluation of [ <sup>18</sup>F]BIT1, the best candidate among this series.

Usually, it is broad basic and clinical research on the involvement of potential imaging targets in brain diseases that strongly support the development of related PET/SPECT radiotracers. However, the review of Cummings et al. about the molecular imaging of opioid receptors (ORs) and opioid-receptor-like receptors (ORL) concludes that, in this field, it applies only to μOR, while there is scant documentation of δOR, κOR or ORL1 receptors in healthy human brain or in neurological and psychiatric disorders. Here, clinical PET research must catch up with the recent progress in radiopharmaceutical chemistry.

With the development of radiolabeled amino acids for PET and SPECT imaging, a completely different set of targets for molecular brain imaging was facilitated: brain tumors. The identification of LATI, the sodium- independent L-type amino acid transporter 1, as a light chain of the CD98 heterodimer which is strongly overexpressed in C6 glioma cells, stimulated the radiolabeling of a great variety of amino acids for the purpose of brain tumor imaging. The review by Werner et al. summarizes the clinical value of a variety of tracers that have been used in recent years, for the following indications: the delineation of tumor extent (e.g., for planning of resection or radiotherapy), the assessment of treatment response to systemic treatment options such as alkylating chemotherapy, and the differentiation of treatment-related changes (e.g., pseudoprogression or radiation necrosis) from tumor progression. It also provides an overview of promising newer tracers for the investigation of these questions. The authors conclude that currently, the best-established PET tracers in neuro-oncology are radiolabeled amino acids targeting L-system transporters.

A thematically related review by Drake et al. on brain tumor imaging by PET is focused on glioblastoma. It includes most recent experimental approaches such as sigma receptor imaging, as well as PET imaging of the programmed death ligand 1 (PD-L1), the ADP-ribose polymerase (PARP) and the mutated form of isocitrate dehydrogenase (IDH). The authors conclude that these new PET imaging targets have the potential to enhance diagnosis, staging, and treatment approaches for glioblastoma.

In summary, I regard this to be an interesting collection of papers to get an overview on radiolabeled molecules which are preclinically and clinically used for molecular brain imaging. Future perspectives are also considered, particularly for neurodegenerative diseases and brain cancer.

> **Peter Brust** *Editor*

## *Review* **PET Radiopharmaceuticals for Alzheimer's Disease and Parkinson's Disease Diagnosis, the Current and Future Landscape**

#### **Bright Chukwunwike Uzuegbunam 1, Damiano Librizzi <sup>2</sup> and Behrooz Hooshyar Yousefi 1,2,\***


Academic Editor: Peter Brust Received: 5 January 2020; Accepted: 17 February 2020; Published: 21 February 2020

**Abstract:** Ironically, population aging which is considered a public health success has been accompanied by a myriad of new health challenges, which include neurodegenerative disorders (NDDs), the incidence of which increases proportionally to age. Among them, Alzheimer's disease (AD) and Parkinson's disease (PD) are the most common, with the misfolding and the aggregation of proteins being common and causal in the pathogenesis of both diseases. AD is characterized by the presence of hyperphosphorylated τ protein (tau), which is the main component of neurofibrillary tangles (NFTs), and senile plaques the main component of which is β-amyloid peptide aggregates (Aβ). The neuropathological hallmark of PD is α-synuclein aggregates (α-syn), which are present as insoluble fibrils, the primary structural component of Lewy body (LB) and neurites (LN). An increasing number of non-invasive PET examinations have been used for AD, to monitor the pathological progress (hallmarks) of disease. Notwithstanding, still the need for the development of novel detection tools for other proteinopathies still remains. This review, although not exhaustively, looks at the timeline of the development of existing tracers used in the imaging of Aβ and important moments that led to the development of these tracers.

**Keywords:** Alzheimer's disease; Parkinson's disease; β-amyloid plaques; neurofibrillary tangles; α-synucleinopathy; positron emission tomography (PET); diagnostic imaging probes

#### **1. Introduction**

Of all the causes of dementia, AD stands in first place and makes up the largest part—about two-thirds—of all differential diagnoses [1–3], and it is the most common form of dementia in persons older than 65 years [4]. Others have vascular dementia, mixed dementia, PD, Lewy body dementia (LBD) or frontotemporal degeneration (FTD) [2]. Although AD and PD present markedly different clinical and pathological features, many mechanisms involved in AD and PD may be the same, such as mutation in genes, the roles of α-synuclein and tau protein aggregates in oxidative stress and mitochondrial dysfunction, dysregulation in the brain homeostasis of iron [5].

The WHO in 2012 named the prevention and control of neurocognitive disorders (mild cognitive impairment (MCI) or Alzheimer's type dementia) a global public health priority. As of 2012, it was estimated that worldwide 35.6 million people are living with dementia. By 2030 this number will double and by 2050 triple [3]. The World Alzheimer Report also in 2018 estimated that there are 50 million people in the world with dementia. This number by 2050 is likely to rise to about 152 million people [2] a projection not far from that made by the WHO way back in 2012.

In the pathogenesis of AD two proteins are implicated β-amyloid peptide aggregates (Aβ) and tau. Based on several scientific evidences, AD is histopathologically characterized by the progressive deposition of Aβ peptides into the interneuronal space [2,6,7]. The pathogenic pathways leading to AD involve several mechanisms which include the dysfunction of cholinergic neurons and the aggregation of tau, however, it has been shown that the amyloid cascade plays a significant role.

The amyloid cascade assumes that the pathogenesis of AD is as a result of a dysfunction in the synthesis and the secretion of the amyloid precursor protein (APP), usually cleaved by the proteases in the secretase family. Normally, the cleavage of APP by α-secretase within the Aβ domain releases soluble APP-α which is non-pathologic, whereas, in pathology, Aβ is generated from APP via successional cleavages by β-secretase followed by the γ-secretase complex, which cuts the γ-site of the carboxyl-terminal fragment of APP producing two major Aβ isoforms: Aβ1-42 and Aβ1-40, which subsequently aggregate to form β-amyloid plaques [8,9]. Aβ1-42 comprises a major part of amyloid plaques owing to its low solubility and tendency to form aggregates with β-pleated sheet structure [9].

Neurodegeneration and neuronal dysfunction are caused by the binding of extracellular Aβ oligomers to the neuronal surface, leading to functional disruption of a number of receptors, finally culminating in dysfunction and neurodegeneration [2,10]. The accumulation of hyperphosphorylated tau protein in neurons, which normally is a microtubule-associated protein (MAP) abundantly expressed in the central nervous system, is another key player in the pathogenesis of AD. As a result of abnormal hyperphosphorylation the protein self-aggregates and forms paired helical filaments (PHF), which leads to the formation of intracellular neurofibrillary tangles, which ultimately block the neuronal transport system [2,11,12].

A definitive diagnosis of AD still requires a histological examination of post-mortem brain sample [13–15]. However, in living patient's cerebrospinal fluid (CSF) biomarkers and positron emission tomography (PET), in combination with several new clinical criteria can assist in the diagnosis [16,17], and for symptomatic patients with familial early-onset AD, it is recommended to undergo clinical genetic testing together with their asymptomatic relatives [18–20].

The European Medicines Agency has presented the measurement of Aβ peptides and total tau protein levels in the CSF as a complementary usable tool in the diagnosis and monitoring of AD [21,22]. Albeit a less expensive method of evaluation, the method is invasive and carries the risks of adverse effects and discomfiture associated with a lumbar puncture [23–25].

Non-invasive modern imaging techniques allow to identify either patients who are at risk of developing AD, and also to monitor disease progression or both [26–28]. Positron emission tomography (PET) imaging especially, which is superior to other imaging techniques in terms of sensitivity, since only picomolar concentrations of the radiotracers are required allows to visualize, characterize and quantify physiological activities at molecular and cellular levels [29,30]. Hence, it may serve as an important diagnostic tool in the field of drug discovery and development, in order to monitor disease progression and the interaction of ligands with their targets.

Aβ is the most studied and first target for the neuroimaging of AD [31], hence it is no surprise that there are already selective PET radiotracers for its imaging. In 2003, Mathis et al. reported the carbon-11 labeled Pittsburgh compound B ([11C]PiB), and the first successful Aβ-selective PET radioligand, which is a derivative of thioflavin (Th-T) an amyloid-binding histological fluorescent dye [32,33].

The discovery of [11C]PiB led to further tracer development of other Aβ tracers. Three of which are already FDA approved and are 18F-labeled [27] (a radioisotope with a relatively longer half-life of 109.7 min [34], in comparison to carbon-11 with a shorter half-life of 20.3 min, a property that logistically limits its use to centers with cyclotron on-site [35]): [18F]florbetaben (Neuraceq) [36]; [18F]florbetapir (Amyvid) [37]; [18F]flutemetamol (Vizamyl) [38].

So far, there are other findings that the density and neocortical spread of NFTs correlate better with neurodegeneration and cognitive decline in AD patients [39–42], in spite of Aβ pathology temporarily preceding tau pathology [27]. Recent evidence further corroborates initial findings of the dominant role

of tau in the pathogenesis of AD [39,43,44], backing this protein as a diagnostic as well as a therapeutic target [45].

Moreover, since apart from AD there are other NDD associated with amyloid pathology, amyloid imaging is not enough to differentiate dementia subtypes [27]. Nevertheless, NFTs are also present in other dementias, like FTD, some neurodegenerative movement disorders like corticobasal degeneration (CBD) and progressive supranuclear palsy (PSP) [45]. More recently, Vanhaute et al. reported that the loss of synaptic density in the medial temporal lobe is linked to an increased tau deposition in AD [46,47]. Hence, a radiotracer, that could quantify NFTs would help to understand the pathophysiology and clinical management not only of AD, but these other NDD. Furthermore, when done in conjunction with amyloid diagnosis, PET imaging of NFTs might provide a way to distinguish between AD dementia (when there are NFTs and Aβ present) and non-AD dementia (when NFTs and Aβ are absent). Furthermore, the application of Aβ imaging is just approved for the exclusion of AD in patients with cognitive impairment but amyloid PET-negative [48]. Also, it is being evaluated as a diagnostic tool for the definition of the preclinical stages of AD [49]. Due to the abovementioned reasons, several academic and industrial groups are currently making efforts to develop tau aggregate tracers, which are not only selective, but also with minimal or no off-target binding [50–53].

The α-synucleinopathies: PD, LBD, multiple system atrophy (MSA) have their pathological hallmark as α-syn aggregates included in Lewy body (LB), Lewy neurites (LN), and glial cytoplasmic inclusions (GCI) in MSA [54–57]. α-Synuclein is a small (140 amino acid residues) highly soluble presynaptic protein that normally exists in a native unfolded state. In PD, there is formation of highly ordered insoluble aggregates known as α-syn fibrils, which are stabilized by β-sheet protein structure [58–61].

The identification of point mutations in the SNCA gene in familial cases of PD nearly 23 years ago first linked α-syn to PD [62], and this was corroborated by the additional discovery that increased genetic copies of α-synuclein in the form of duplications and triplications of the SNCA gene are enough to cause PD; the higher gene copy, the earlier the age of disease onset and the more severe the disease [63–65]. More recently, further investigation into the genetic aspects of the disease culminated in genome-wide association studies (GWAS), and candidate gene association studies which have repeatedly validated that statistically relevant signals linked to PD are common variants near the SNCA, LRRK2, MAPT and low-frequency coding variants in GBA (glucocerebrosidase) genes [66]. Moreover, in GWAS so far, not less than 41 risk loci for PD have been identified [67,68]. Even in the sporadic forms of the disease, α-syn as a candidate risk gene has shown significant associations between variation within the SNCA gene and a higher risk of developing PD [69].

It has been known for some time now based on fairly strong evidence that the motor phase of classical PD occurs after a premotor period that could last for a considerable number of years if not decades [70]. Before the appearance of motor symptoms, at least 50% of substantia nigra (stage 3 of the Braak staging) cells have to be lost [71,72] and likely a loss of a higher percentage of dopaminergic nerve endings in the putamen [73]. Based on the findings of Braak et al., there are 6 stages in which the deposition of α-syn in LBs and LNs occurs sequentially and additively [74]. Overall, it is evident that pathophysiological changes in the central nervous system in PD involves the abnormal deposition of α-syn occurs early in PD, hence the earliest definition and most precise detection of premotor PD should be based on the imaging of aggregate α-syn, not dopaminergic alterations.

Despite the high abundance of α-syn in the nervous system, where it constitutes 1% of all cytosolic proteins [75], the amount of α-syn aggregates, however, in LBD and MSA brain is 10-fold or lower than that of Aβ in AD brain, and in advanced cases in the range of 50–200 nM in brainstem and subcortical regions, and moreover, they typically have a small size, which complicates detections [76,77].

Unlike Aβ, but similar to NFTs, LBs are intraneuronal and GCI are intraglial, hence any tracer for the detection α-syn must readily pass through the blood-brain barrier (BBB), and subsequently the cell membrane to access its target [77,78]. Unfortunately, due to the structural similarity of β-pleated sheets amongst different species of amyloid fibrils, and the colocalization of α-syn aggregates with other aggregating amyloid proteins like Aβ plaque and tau fibrils tracer, selectivity for α-syn aggregates over the others is a desired quality. This explains why non-selective ligands are more common than selective tracers [79–82].

Generally, good PET radiotracers for brain amyloid imaging should have the qualities prerequisite for successful central nervous system ligands [83,84]. A good brain penetration via passive diffusion, relatively small molecular weight (< 700 Da), moderate lipophilicity 1–3 at physiological pH (7.4), lack of P-glycoprotein substrate activity, lack of BBB permeable radioactive metabolites or intracerebral radiometabolites, etc. Most importantly, they should with high affinity selectively and reversibly bind to targets in the brain. Target selectivity an important trait depends on factors such as the relative affinities of the tracer to target (specific binding) and non-target (non-specific binding) sites, its brain distribution and the relative concentration of the binding sites. Both target and non-target binding sites should be considered when developing a brain tracer [77,81,82,85,86].

Additionally, a slow and reversible off-rates coupled (koff) with relatively high on-rates (kon), which is reflected by an equilibrium dissociation constant (Kd) in the range of 1 nM. A low Kd value in the nanomolar (nM) range could guarantee that the radioligand-amyloid complex remains intact long enough for a washout of non-specifically bound tracers to occur, hence allowing good signal-to-noise contrast. It is also needed especially when dealing with short-lived PET radioisotopes like 11C with a half-life of 20.3 min and 18F half-life 109.8 min. A standard uptake value (SUV) in the brain > 1.0 within a few min of intravenous injection is also required. Large molecules, antibodies, and nanobodies can cross the BBB, however they are unable to attain an SUV value > 1.0 a few min post-injection (p.i), and this has been a disqualifying criterion for large ligands labeled with short-lived radioisotopes [81,85,86].

#### **2. PET Imaging Agents for the Diagnosis AD and PD**

#### *2.1. PET-Tracers for the Imaging of A*β *Plaques*

#### 2.1.1. First Generation of Aβ PET Tracers

#### Benzothiazole (BTA) Derivatives

The development of amyloid-specific imaging compounds is based mostly on conjugated dyes like Th-T (Figure 1) and Congo red, that are used in postmortem AD brain sections for the staining of plaques and tangles [87–90]. The synthesis of the hundreds of the derivatives of the latter by the Pittsburgh group gave rise to a series of pan-amyloid imaging agents that showed nanomolar binding affinities for Aβ, tau, α-syn, and prion aggregates. Notwithstanding, a number of these compounds ionize at physiological pH, and for this reason did not achieve high brain uptake ( > 1 SUV) a few min post intravenous injection [32,91].

**Figure 1.** Structures of thioflavin-T, [11C]PiB, and the FDA approved Aβ-PET tracers: [18F]florbetaben, [ 18F]florbetapir, and [18F]flutemetamol.

The examination of the derivatives of Th-T derivatives followed: making the dye neutral by the removal of the methyl group attached to the benzothiazole ring via the nitrogen atom of the ring, hence the positive charge on the benzothiazole ring gave rise to compounds (known as benzothiazole anilines or BTAs) with improved lipophilicity, [11C]6-Me-BTA-1 (Figure 2) being the best in the series. It was 6-fold more lipophilic, and readily crossed the BBB in brains of rodents, and showed 44-fold more affinity for synthetic Aβ fibrils (Table 1) than did Th-T [92,93].

**Figure 2.** Structures of the predecessors of FDA approved Aβ-PET tracers: [11C]6-Me-BTA-1, [11C]SB-13, [ 18F]FMAPO.

Further manipulation of the benzothiazole ring by derivatizing the C-6 position and varying the degree of methylation of the aniline nitrogen gave a series of ligands with high affinity for Aβ fibrils. Of these radiotracers, the monomethylated-aniline derivative ([11C]PiB [11C]6-OH-BTA-1 (Figure 1), was selected (which will be referred to as just PiB throughout the paper). It showed a combination of favorable pharmacokinetics as PiB, the highest brain clearance 5 times faster than at 30 min and a high binding affinity to Aβ plaques approximately 207-fold than Th-T [94] (Table 1), with a very low binding affinity to aggregated tau, with a ratio of tau-to-Aβ ((Kitau/KiAβ) greater than 100-fold [33,95–97].

Clinical study with PiB showed that AD patients retained PiB in areas of association cortex known to contain large amounts of amyloid deposits [33]. Further clinical studies to confirm if there is abnormal binding of PiB in clinically healthy individuals showed that PiB-PET not only was able to detect Aβ deposits in AD patients but also in some nondemented patients, hence suggesting that amyloid imaging might be useful in the detection AD in its preclinical stages [98]. Additionally, it was confirmed that there is a direct correlation of the retention of PiB in vivo with region-matched quantitative analyses of Aβ plaques in the same patient, upon post-mortem examination of clinically diagnosed and autopsy-confirmed AD subjects [99]. This too additionally validated PiB-PET as a method for evaluating the amyloid plaque burden in AD subjects [33].

In an experiment carried out by Serdons et al. it was discovered that more than 80% of the tracer remains intact 60 min p.i [100,101]. The radiometabolites of PiB found in animal and human blood, due to their high polarity did not easily pass through the BBB [94,100]. One of the identified radiometabolites 6-sulfato-O-PiB, and others produced in rat brain, built up over time and complicated pharmacokinetic analyses [95,102]. Fortunately, the intracerebral metabolism of PiB is limited only to rats and was not observed in mice, humans, and other nonhuman primates [95].

The success of PiB for in vivo imaging of Aβ plaque deposition led to the development of an 18F analog, which would perform similarly. The development of 18F-labeled radiotracers for the imaging of amyloid deposits in AD was on the basis that, as previously mentioned, carbon-11 with which PiB was labeled has a half-life 20.3 min, and this limits its use to PET centers with cyclotron on-site and with experience in 11C-radiochemistry [33,36].

A variety of structural analogs were developed and evaluated both in vitro [103] and preclinically, out of which flutemetamol also known as [18F]GE067 ([18F]3- F-PiB) (Figure 1) was selected [104]. In vivo studies in rats and mice showed that it has similar pharmacokinetics as PiB. They both readily entered the brain, however, flutemetamol which is more lipophilic was washed out more slowly from the brain approximately 1.4 times slower (Table 1), especially from the white matter [105].

Initial human studies, in which flutemetamol and PiB were compared in AD and control subjects, the former showed similar uptake and specific binding attributes as PiB [104]. A phase-III trial

demonstrated that it is safe with high specificity and sensitivity for the in vivo detection of brain Aβ density [106,107]. It was approved by the FDA in 2013 [108].

#### The Stilbene and Styrylpyridine Derivatives

The discovery of [3H]SB-13, a stilbene derivative which showed a high binding affinity to postmortem AD brain homogenates [109], led to subsequent labeling with carbon-11 to afford [ 11C]SB-13 (4 methylamino-4- -hydroxystilbene) (Figure 2). The tracer displayed a good brain uptake and brain clearance (Table 1) [110]. In vivo human PET-imaging it displayed properties similar to PiB in discriminating between AD and non-AD patients [111].

The similarities between PiB and SB-13 in addition to their similar biological properties are also in their chemical structures: the presence of a highly conjugated aromatic ring with an electron-donating group (N-methylamine (-NHCH3) or hydroxyl (-OH)) at the end of the molecule and the relative planarity of both ligands [90].

Early attempts at the development of 18F-labeled SB-13 was unsuccessful, due to the high lipophilicity and high nonspecific binding in the brain shown by [18F]SB-13 derivatives with a fluoroalkyl group on either ends of their structures. In order to reduce the lipophilicity of the ligands, the stilbene scaffold was further modified by the introduction of different functional groups. Based on in vitro and in vivo biological assays a NH-CH3 derivative [18F]FMAPO, with a 2-fluoromethyl-1,3-propylenediol group tethered to the phenol end of molecule (Figure 2) was selected for not only exhibiting a selectivity and specific binding to Aβ plaques in AD brain homogenate binding studies but also for showing a higher brain penetration in 2 min, which was nearly three times higher than that of flutemetamol in 5 min (Table 1). Although it displayed a slower washout than the latter, at 60 min p.i. the concentration in the brain was less than 1%ID/g [103,112].

In order to circumvent the complication of in vivo metabolism, which might result due to the presence of a chiral center in the fluorine containing side chain, another series of stilbene derivatives were synthesized with polyethylene glycol (PEG) units of different lengths (n = 2–12) tethered to the 4- -OH group, with 18F attached at the end of PEG side-chain. This also provided a way to maintain a small molecular weight, adjust lipophilicity and facilitate a simple 18F-labeling by nucleophilic substitution. Structure-activity relationship (SAR) studies showed that high binding affinity was maintained when n < 8, and from 8 and above there was a significant reduction in binding affinity. There was a noticeable decrease in brain penetration as shown by in vivo biodistribution studies when n > 5 [113–115], perhaps partly due to increased molecular weight and total polar surface area (tPSA).

Of the four ligands which performed well in in vitro and in vivo assays, florbetaben (Figure 1) also known as AV-1, or BAY94-9172 with n = 3, was selected. Although the tracer did not have the highest affinity for Aβ in comparison with its structural analogs or the fastest washout rate (Table 1) from the brain of healthy mice [114], it, however, showed selectivity for Aβ and non-appreciable binding to NFTs, Pick bodies, LBs and GCIs [112]. Furthermore, binding to postmortem cortex of subjects with FTD or postmortem brain tissue from other NDDs like tauopathies and α-synucleinopathies was not observed [114]. With no observable effects at 100x the expected human dose in preclinical toxicity studies in a different animal species, florbetaben was deemed suitable for human studies [116]. In 2014, it was approved by the FDA [117].

In order to obtain an Aβ tracer with improved in vivo biological properties of targeting Aβ plaques, so that a high signal to noise ratio is quickly and more efficiently achieved, some critical and competing factors were taken into consideration: initial brain uptake, washout from non-afflicted brain regions, in vivo metabolism, and optimal time in the accomplishment of the highest target-to-non-target ratio. For this reason, the stilbene ring was further explored. The fluoropegylation discussed above was extended from stilbene to styrylpyridine series. This was achieved by exchanging one of the stilbene benzene rings for a pyridine ring. This led to the development of florbetapir also known as [ 18F]AV45 [118]. It displayed 2-fold more binding affinity to Aβ in postmortem AD brain homogenates than florbetaben. Nevertheless, it showed a slightly lower initial uptake and washout rate from the brain of healthy mice than florbetaben (Table 1) [114,119].

An initial clinical trial with a tertiary amine derivative, which was similar to florbetapir but for the dimethylation of the aniline nitrogen suggested lower than expected brain uptake, probably due to a fast in vivo metabolism by N-demethylation. Of all the evaluated ligands, faster brain kinetics was exhibited more by florbetapir, and it also displayed an excellent brain uptake and washout in humans. The signal to noise ratio in the brain approaches an optimal level in 40–60 min post intravenous injection. In vitro metabolic stability assay also demonstrated that it is more stable towards microsomal degradation than florbetaben [115,118].

In AD patients, florbetapir from 30 min p.i. showed a clear separation between cortical and cerebellar activity, hence making it possible to start brain PET scan 30–50 min p.i. [120]. Significant elevations of tracer uptake in several brain regions of AD patients in comparison with controls were observed upon visual evaluation and analysis using semiquantitative methods. Results from phase III clinical trial showed a distinct correlation between the distribution of Aβ and florbetapir PET images at postmortem examination. Furthermore, no serious side effects were recorded in any of the clinical trials of the tracer [121]. It was approved by FDA in 2012 [31,122].

#### 2.1.2. Second Generation of Aβ PET Tracers

#### Benzofuran, Benzoxazole and Imidazobenzothiazole Derivatives

Other notable Aβ tracers include flutafuranol, also known as [18F]AZD4694 ([18F]NAV4694) (Figure 3) a benzofuran derivative, developed by researchers at AstraZeneca in Sweden [123]. Its development, amongst other second generation of 18F-labeled Aβ imaging agents [124] was spurred by the report that flutemetamol and florbetaben, have high level of non-specific white matter retention [116,125,126], which could be a limitation in situations when insoluble Aβ levels are low, due to a spillover effect of radioactivity to adjacent cortical regions from nonspecific binding in white matter. Hence, they may not be useful for correct mapping of Aβ plaque load in low-density regions and in prodromal phases of AD.

**Figure 3.** Structures of second generation Aβ-PET tracers: [18F]AZD4694, [18F]MK-3328, [18F]AD-269, [ 18F]FIBT.

Using the intravenous cassette dosing technique to compare the pharmacokinetics of flutafuranol and flutemetamol, it was seen that they were both readily taken up in the brain tissue and washed out of the brain normal rats between 2 and 30 min, but with less than 10% of concentration of flutafuranol at 2 min remaining at 30 min, a time point at which flutemetamol still had up to 28% of the initial concentration at 2 min (Table 1) [103]. With its fast binding kinetics, it could perform better than other Aβ tracers, like PiB, which display, based on time-activity curves, slower kinetics with a blunt peak of specific binding accompanied by a slower decline [95,127]. Consequently, its rapid binding kinetics makes quantification using data based on short acquisition possible. Furthermore, using the cerebellum as a reference region in approaches like reference Logan, valid estimates of Aβ binding could be easily acquired [128]. It is presently in its phase III of clinical trial for the evaluation of its efficacy and safety for the detection of cerebral Aβ in comparison with postmortem histopathology [129,130].

A benzoxazole derivative [18F]MK-3328 (Figure 3), which was selected amongst four other fluoroazabenzoxazoles owing to its favorable kinetic profile, shown in rhesus monkey PET studies, a relatively low binding potential in white matter and cortical grey matter, which is approximately 2× lower than that of florbetapir, a relatively lower lipophilicity at log D 2.91, in comparison with an analog [18F]AD-269 with similar properties, but more (1.21 fold) lipophilic (Table 1).

In autoradiography studies, it was observed that in an AD patient brain slice that MK-3328 showed punctuate, displaceable binding in the cortical gray matter, with no noticeable binding in the cerebellum [131]. Investigation of the tracer in healthy human volunteers and AD subjects was also being carried out at the time until the premature termination of the clinical trial after the completion of phase 1 of its clinical trial [132].

The best imidazobenzothiazole derivative [18F]FIBT (Figure 3) was reported by the Yousefi et al. group and it was described as the first high-contrast Aβ-imaging agent on par with florbetaben (Figure 4). It also displayed excellent pharmacokinetics, selectivity and high binding affinity to Aβ fibrils in vitro and in vivo comparable to the gold standard PiB [133–135].

Their results also showed that FIBT has a better pharmacokinetic profile and specific binding affinity to Aβ than florbetaben in transgenic mice. This could be expected from a tracer with >300-fold selectivity for Aβ in comparison to the other amyloid protein aggregates a Ki >> 1000 nM to recombinant tau and Ki >> 1000 nM to α-syn aggregates [114,136]. Further investigations of the tracer in human subjects are however yet to be carried out [133,135].

**Figure 4.** Exemplary sagittal PET images of the FDA approved Aβ PET-tracers of Alzheimer's disease patients with other select featured tracers, [11C]PiB, [18F]Florbetaben, [18F]Flutemetamol, [ 18F]Florbetapir, [18F]Flutafuranol, and [18F]FIBT (reproduced with permission as agreed by Newlands Press Ltd. [135]).


*Molecules* **2020**, *25*, 977

#### 2.1.3. The Clinical Utility and Consequences of Clinically Approved PET-Aβ Radiotracers

Since the clinical approval of the abovementioned three FDA approved PET-Aβ tracers as diagnostic tools for the detection of neuritic (Aβ) plaques in live patients, there have been studies to determine their clinical usefulness in the diagnosis AD. These studies have been subsequently and specifically well-reviewed by Kim et al. [137], Barthel et al. [138], Chiotis et al. [139].

In general, the studies have showed that the use of the Aβ-PET tracers led to a moderate to significant change in diagnosis, diagnostic confidence [140–145], and had a substantial impact on change in the treatment and management plan of AD [137,144,146]. It is likely that the new generation of Aβ-PET tracers with improved pharmacokinetics will allow for improved signal-to-noise ratio, hence will be more suitable for the quantification of disease progression and therapeutic monitoring.

#### *2.2. PET-Tracers for the Imaging of Tau Aggregates*

As mentioned earlier, the tau protein plays a key role in the pathogenesis of AD [2,6,11,12]. The predominant aggregation of certain MAPT (tau gene) isoforms, either the 4-repeat (4R tau) or the 3-repeat (3R tau) isoforms have been widely described in tauopathies. So, in addition to the already mentioned properties every CNS tracer should possess [65,67,68,73,74], tau tracers must also address 3R and 4R tau deposits. 3R and 4R tau proteins are the classifications of the 6 tau isoforms according to their tubulin-binding domains [147,148]. In a normal brain, there are equal amounts of the 3R and 4R tau proteins, as well as in AD. An imbalance in tau ratio can lead to abnormal tau accumulation and lead to NDD as in tauopathies. For instance, there is an ample amount 4R tau in PSP, CBD, and argyrophilic grain disease, in contrast, there is an abundance of the 3R tau in Pick's disease (PiD) [149]. Furthermore, tau tracers should also be able to bind to different tau folds, all of which will facilitate the detection of tau pathology in both AD and non-AD tauopathies [150,151].

The identification of lead compounds for the imaging of Aβ proteinopathies has been relatively easier since most β-sheet binding ligands have a high affinity for Aβ fibrils, with which NFTs coexist in AD and both are colocalized in the gray matter structures [42,152]. In spite of controversy surrounding the subject, PHFs predominantly found in NFTs in in vitro experiments suggest a β-sheet structured core similar to that characteristic of Aβ and α-syn fibrillar aggregates [42]. There are recent reports that there are α-syn containing aggregates present in AD [79,153]. Therefore, tau PET tracers should be selective for tau aggregates over these aggregates as well.

In AD, there is a distinct difference in the concentration of Aβ relative to tau aggregates. The concentration of Aβ is approximately 5–20 times higher than that of tau aggregates [154]. In spite of this inequality in quantity, there is however a clear-cut regional pattern of Aβ and tau deposition in the neocortex. The frontal cortex has the highest concentration of Aβ aggregates, while the temporoparietal cortices have the highest concentrations of tau aggregates. Different distributions of tau aggregates are also found in the different phenotypes. Although this has its own merits as it will facilitate differential diagnosis of tauopathies, it means that it is unlikely a single tau PET tracer could bind to the whole spectrum of tau polymorphism [42]. In this review, some select selective tau tracers already evaluated in human subjects will be examined, together with other notable tau tracers.

#### 2.2.1. First Generation of Tau-PET Tracers

#### The Arylquinolines

The THK-compounds (Figure 5) were as a result of the structural modification of the lead compounds BF-158 and BF-170, arylquinoline derivatives. Even though in vitro fluorescence binding affinity assay data and neuropathological suggested that they are good tau ligands, they showed poor selectivity over amyloid plaques, and furthermore were unable to bind to tau present in non-AD tauopathies. However, autoradiographic studies in AD brain section showed an uptake BF-158 in brain regions which were NFT-rich. Biodistribution studies, analyzed using HPLC with a fluorescence detector, showed a good uptake BF-158 (11.3% ID/g at 2 min p.i.) of BF-158 in the brain of normal mice but a slow washout with only 27.4% of the concentration at 2 min washed out at 30 min, which suggested a high unspecific binding (Table 2). In contrast BF-170 performed better in the biodistribution studies with good brain uptake, as well as a faster washout at 30 min than BF-158 [155] (Table 2).

**Figure 5.** Structures of the first generation tau-PET tracers: BF-158, BF-170, [18F]THK-523, [ 18F]THK-5105, [18F]THK-5117, [18F]THK-5317(17), (S)-[18F]THK-5117 ([18F]THK-5351), [11C]PBB3, [ 18F]Flortaucipir (AV-1451, [18F]T807), [18F]T808.

**)@7 \$9**

**>**

Structural modification of BF-170 led to the development of [18F]THK-523 (Figure 5) a [ 18F]fluoroethoxy derivative. The introduction of an alkylether in the C6 position of the arylquinoline structure improved its affinity and selectivity for tau aggregates relative to BF-170 [156]. However, competition studies showed that it has a relatively low affinity for recombinant tau fibrils (Ki 59.3 nM) and even lower for PHF in AD brain homogenates (Kd 86.5 nM) than synthetic heparin-induced tau polymers (HITP) Kd 1.67 nM (Table 2), an evidence of the inadequacies of synthetic tau preparations, which fails to completely replicate native tau aggregates in vivo [156,157].

Notwithstanding, it performed betterin vivo than it did in vitro. In comparison to healthy controls, it showed higher cortical retention in AD subjects and was distributed in the brain in accordance with reported histopathological brain distribution of PHF in AD. Unfortunately, due to its high white matter retention, a clear visualization of PET scans was not possible, and for this reason, it was not further developed [158].

Introduction of a secondary alcohol in the fluoroethoxy chain in BF-170 and the monomethylation [ 18F]THK-5117 and dimethylation [18F]THK-5105 (Figure 5) of the aniline moiety gave tracers with higher in vitro affinities (Kd) for both synthetic HITP tau fibrils and for human AD-PHF tau aggregates than [18F]THK-523: they showed a 16-fold and 32-fold increase in affinity for human AD-PHF tau aggregates in comparison to their direct predecessor [18F]THK-523, a higher in vitro selectivity for tau versus Aβ; a coincidence with Gallyas-Braak staining and immunoreactive tau staining in autoradiography staining of human AD brain sections but not with the distribution of PiB and an good initial brain uptake and washout in normal mice than [18F]THK-523 [157] (Table 2).

An improved selectivity could be due to the secondary alcohol present, a polar terminus in the molecules. Likewise the presence of a secondary amine in [18F]THK-5117 and a tertiary amine in [18F]THK-5105 seemed to be behind the enhancement of tau affinity and better pharmacokinetic profile [35].

However, the N-dimethylation of [18F]THK-5105 appeared to be its undoing: due to its relatively higher lipophilicity (1.3x) than [18F]THK-5117 (Table 2), it showed in vivo nonspecific binding in the brainstem, thalamus and subcortical white matter, which hinders interpretation. Notwithstanding, it was still able to differentiate between AD patients and healthy control in its first-in-human PET studies. Its distribution in the mesial and lateral lobes of AD patients is in accordance with the reported NFT distribution in AD brain. Nevertheless, due to its inadequacies in comparison to other known tau-PET tracers, it was not used further [159,160].

On the other hand, first human PET studies with [18F]THK-5117, demonstrated that it has faster kinetics and better signal to noise ratio than seen in [18F]THK-5105, in comparison to which it is less lipophilic. Clinical studies have been conducted with the (*S*)-enantiomer [18F]THK-5317, owing to its better signal-to-noise ratio and pharmacokinetics than the (*R*)-enantiomer, a trait observed in the quinoline derivatives [159,161–163]. One of the shortcomings of [18F]THK-5117 and its S-enantiomer [ 18F]THK-5317 is their significant white matter binding, which might be owing to binding to β-sheet structures of myelin basic protein [164].

In order to reduce white matter binding a feature common among 18F-labeled amyloid tracers, a structural modified (S)-[18F]THK-5117 was developed, the phenyl ring was replaced with a pyridinyl ring, which made [18F]THK-5351 more hydrophilic [164]. It not only displayed a quicker white matter washout (lower white matter retention) and higher specific binding to AD tau-associated regions of interest than [18F]THK-5317, but also its retention correlated with extra-hippocampal sub-regional atrophy rather than hippocampal subfields, proffering hence different underlying mechanisms of atrophy in early AD. Another remarkable advantage it has over other tau tracers was the lack of significant retention in the choroid plexus or venous sinus, which could probably lead to a spill-in of tracer signals into the brain [165,166].

Unfortunately, it has been reported to have high affinity to monoamine oxidase-B (MAO-B) (an isoform of monoamine oxidase whose function is to catalyze the oxidation of monoamines [167,168]) in contrast with [18F]THK-5117, and also showed a greater off-target binding in the midbrain, thalamus and the basal ganglia [169,170].

#### The Phenylbutadienylbenzothiazoles (PBB)

Following the observation that ligands with a π-electron-conjugated backbone longer than 13Å showed affinities for pathological inclusions in a several tauopathies Maruyama et al. investigated the affinities of a series of compounds with a different structural dimension to tau aggregates and concluded that a core structure with specific distance from 13–19 Å contributes to affinity for non-AD inclusions. Additionally, since ligands with a slender and flat backbone have the ability to transverse and attach to channel-like channels in β-pleated sheets they developed a class of compounds phenyl/pyridinyl-butadienyl-benzothiazoles/benzothiazoliums (PBBs) [171]. They are structural analogs of fluorescent amyloid dye Th-T, with an all-trans butadiene bridge between the

aniline and benzothiazolium moieties. Interestingly the resulting tracers were also able to detect tau inclusions in non-AD tauopathies like CBD, Pick's disease and PSP [172].

Amongst a series of analogs, [11C]PBB3 (Figure 5) was selected as the best candidate with an affinity Kd for HITP 2.55 nM and nearly 50 folds selectivity for tau versus Aβ fibrils [35,172] (Table 2). Based on preclinical findings in mice, the tracer was further evaluated in humans. It showed in comparison with control accumulation in the medial temporal region of AD subjects. Its distribution in AD human brains differed from that of PiB, suggesting minimal nonspecific binding to white matter, although in both controls and AD brains it accumulated in dural venous sinuses. The use of the tracer in a CBD patient showed its retention in the basal ganglia, hinting that it could be useful for the imaging non-AD tauopathies additionally [172].

The compound although it seemed to be a likely candidate for the in vivo imaging of tau pathology, regrettably had some in vitro and in vivo instability problems. The in vitro instability was due to its photoisomerization tendencies: the quick interconversion of E/Z isomers in the presence of light. Although, this can be suppressed by shielding it from light during radio- and chemical synthesis it still is an inconvenience. In vivo, it gets quickly metabolized in mice and humans, with 2% remaining unchanged in mice at 1 min p.i. and 8% at 3 min p.i. in humans. Although this radiometabolite is polar, in mice it still made image analysis difficult [173]. It also displayed off-target binding in the basal ganglia, the choroid plexus and the longitudinal sinus [172]. Nevertheless, a new fluorinated PBB compound [18F]PM-PBB3 (Figure 6) has been developed and is being clinically investigated to find out if there would be an improvement in the shortcomings of [11C]PBB3 [51].

**Figure 6.** Structures of the selected second-generation tau-PET tracers. [18F]GTP1, [18F]PM-PBB3 (APN-1607), \*9, [18F]MK-6240, \*12, [18F]RO-948 (RO6958948), [18F]PI-2620, [18F]JNJ64349311(JNJ311).

Interestingly, in in vitro fluorescent study using postmortem DLB and MSA brain sections PBB3 was colocalized on α-syn in LBs, LNs, and GCIs. In contrast, autoradiographic labeling with [ 11C]PBB3 at 10 nM only showed significant binding in MSA cases in regions with a high density of GCIs in the absence of tau or iron deposits. Since the maximum concentration of [11C]PBB3 in human PET scans is roughly 10 nM as presented by Koga et al., it means that α-syn is only detectable by [ 11C]PBB3 in MSA patients with a high density of GCIs [174]. A later in vivo human PET study on MSA patients by Perez-Soriano et al. was consistent with the work carried out by Koga et al., that [ 11C]PBB3 binds to α-syn [175,176].

#### The Carbazole and Benzimidazole Derivatives

A screening campaign at Siemens MI Biomarker Research led to the discovery of these classes of lead series [177]. Further optimization led to the development of flortaucipir (AV-1451, [18F]T807) and [ 18F]T808 (AV-680) (Figure 5). They both have sufficient affinity for tau (AD-PHF) 14.6 nM and 22nM based on a Scatchard analysis of autoradiography staining of human PHF-AD brain sections, with Kd(A<sup>β</sup>)/Kd(tau) 25 and 27 respectively, meaning a higher selectivity of tau aggregates over Aβ fibrils (Table 2). The binding of [18F]T808 to only one type of binding site of the tau aggregates as seen from the degree of linearity in its Scatchard plot further confirmed its selectivity for tau aggregates over Aβ. Most importantly, they have minimal white matter binding and a good pharmacokinetic profile (Table 2) [178,179].

Initial PET scans of flortaucipir in controls and subjects with AD and mild cognitive impairment demonstrated an accumulation of the tracer with a distinct increasing neocortical distribution in tandem with the severity of dementia [180] according to the known mode of spread of PHF in the brain in agreement with Braak's staging [181]. In other tauopathies such as PSP and CBD, it was shown in a head-to-head comparison of [11C]PBB3 and flortaucipir, that the former binds more avidly to neuronal and glial tau lesions relative to the vague binding of flortaucipir [182]. Similar to [ 11C]PBB3, it is suspected that flortaucipir significantly binds to α-syn in the posterior putamen MSA patients. This, however, is not consistent with in vitro autoradiography results, which so far has proven otherwise [183,184].

First-in-human PET studies with [18F]T808 showed similar results as flortaucipir, but with more rapid kinetics. As early as 30 min p.i. [18F]T808 images stabilized, but flortaucipir SUVR values after 80 min still fluctuated. Notwithstanding, flortaucipir was selected over [18F]T808 for clinical development, because of the metabolic defluorination observed in some cases, and the significant accumulation of fluorine-18 in the skull especially in late time points, that could confound PET images. This prevented further in vivo use of the tracer [185].

Off-target binding has been seen in flortaucipir PET studies in the meninges, striatum, choroid plexus and midbrain. In the analysis of autopsy brain samples, it was found out that flortaucipir also binds to vessels, iron-associated regions, substantia nigra, the leptomeningeal melanin and calcifications in the choroid plexus [186]. Another important off-target of flortaucipir is to both isoforms of the MAO enzyme [167,168,187,188]. Furthermore, there was difficulty in quantification due to the fact that it does not reach a steady-state during a typical imaging duration [189,190].


**Table2.**BindingaffinitiesandPharmacokineticsoffeaturedfirst-generationtauPET-tracers.

The log P values are the partition coefficient (octanol/water) or log D partition coefficient (octanol/PBS) reported in the respective publications. Selectivity tau vs Aβ: EC50(Aβ)/EC502 Kd(Aβ)/Kd(tau). NA: data not available. 3 Autoradiographic binding to plaque- and tangle-rich regions in AD brains.

#### *Molecules* **2020** , *25*, 977

#### 2.2.2. Second Generation of Selective Tau Tracers

Even though several goals were achieved with the first-generation selective tau tracers like improvement in affinity to both 3R and 4R tau deposits, selectivity of the tracers to tau aggregates versus Aβ plaques, and pharmacokinetics, there remains still the problem of lack of selectivity over other protein aggregates, and brain contents: subcortical white matter accumulation, in conjunction with off-target binding especially to MAO-B enzyme in the basal ganglia. Findings in which the THK-radiotracers and flortaucipir have been implicated following in vitro assessments [169,170,184,188]. Clinical validity could be limited in tauopathies where the accumulation of tau is expected in regions with a high concentration of MAO-B, like in PSP and CBS.

There has, furthermore, also been mounting evidence that the binding of certain tracers such as flortaucipir and 18F-labeled THK tracers is not only limited to tau deposits but to other protein deposits, like TDP-43 (transactive response DNA binding protein 43 kDa) predominantly present in patients with semantic dementia. Flortaucipir as well as [11C]PBB3 showed in vivo binding in patients expected to have α-synuclein deposits [174–176,183,184].

Consequently, efforts are being made by various pharmaceutical companies and research institutes to optimize the binding selectivity and enhance the pharmacokinetic profile of tau PET tracers. Hence, more focus will be paid in this section to improvements in the pharmacokinetic profile and specificity of the new tracers in comparison to their predecessors.

#### Optimized First Generation Tau Tracers

As mentioned earlier [185], [18F]T808 had a propensity to metabolic defluorination, which led to the selection of flortaucipir (AV-1451, [18F]T807) over it, despite its faster kinetics (Table 2). For this reason, it was deuterated to improve its in vivo stability to defluorination, which resulted in the development of [ 18F]GTP1 (Figure 5). This modification prevented the accumulation of free 18F-flouride in the skull in clinical PET study, in which it also distinctly differentiated AD subjects from healthy controls [189,192,193].

In addition to its low nanomolar affinity to tau aggregates and excellent selectivity to Aβ plaque (Table 3), it was reported to bind to non-AD tau aggregates. It also showed no off-target binding especially to MAO-B. Both preclinical and clinical in vivo kinetic studies showed that the tracer has a good pharmacokinetic profile which allows imaging some minutes earlier than flortaucipir [190,194]. Further investigations however still need to be carried out to properly compare these two tracers [193].

An introduction of fluorine-18 in the structure of the first-generation tracer [11C]PBB3 gave rise to [18F]PM-PBB3 (APN-1607) (Figure 6). In human subjects, it showed in less than 5 min a peak ~2.5 [ 18F]PM-PBB3 SUV in the brain. It has less off-target signals in the basal ganglia than [11C]PBB3, and a greater signal-to-background ratio. It showed no significant off-target binding in the basal ganglia and thalamus. Furthermore, it did not show radiometabolites in the brain as did its predecessor [ 11C]PBB3 [192,195]. Phase 0 of its clinical evaluation was completed not so long ago in 2018 [196].

A structurally modified version of flortaucipir whose inadequacies were already discussed [186,187,197], [18F]RO-948 (RO69558948) (Figure 6) was developed and selected from three lead compounds. Of the selected three which also displayed good brain uptake, fast brain clearance, high affinity for NFT (Table 3) and excellent selectivity against Aβ plaques in AD brain tissue, lower affinity for MAO-A and MAO-B in comparison to [18F]T807 and [18F]THK-5351, and based on preclinical binding study RO-948 was selected for further development. This was because in comparison to the other analogs it displayed better pharmacokinetics and metabolic properties both in mice and non-human primates. Moreover, it showed a better signal-to-background ratio than the others in AD patients. Notwithstanding, three of them gave results from their first-in-human study, which were consistent with preclinical data [198–200].

Upon the discovery that affinity for MAO-A is significantly attenuated and high affinity for aggregated tau improved in the presence of pyrrolo[2,3-b:4,5-c']dipyridine core structures in comparison to pyrido[4,3-b]indole core structure a series of fluoropyridine regioisomers were developed from which the 4-pyridine regioisomer [18F]PI-2620 (Figure 6) a regioisomer of RO-948 was selected. In AD brain homogenate competition assays, it demonstrated a high affinity for tau deposits pIC50 8.4 nM (Table 3), and a superior binding to both 3R and 4R tau aggregate folds in self-competition experiments using recombinant K18 fibrils (representing 4R tau pathology) as well as human PSP and PiD brain homogenates.

Besides, it is selective over Aβ and has no off-target binding towards either MAO-A as [ 18F]RO-948 or MAO-B as flortaucipir, and furthermore showed low off binding in brains of non-demented controls, with rapid and complete washout. It also showed selective binding to pathological tau present in Braak I, III and V human brain sections in autoradiography experiments. In autoradiography studies it also showed to tau aggregates/folds in PSP brain sections, which of course has been controversial, since many tau tracers has been reported not to be bind to tau deposits in PSP in autoradiography experiments. However, off-target binding was observed in the pars compacta portion of the substantia nigra in human brain sections, consistent with the affinity of some tau tracers like flortaucipir, [18F]MK-6240 to melanin-containing cells [52,53,201].

Clinical data are needed to confirm the usefulness of [18F]PI-2620 in non-AD patients. Nonetheless, it is presently being examined in several clinical trials in order to establish its pharmacokinetic profile in humans, and decide its application in in vivo PET-imaging of tau aggregates/folds both in non-AD and AD tauopathies [53].

#### The Azaindole-Isoquinoline and Naphthyridine Derivatives

Following an SAR study, an azaindole-isoquinoline derivative was developed. The study showed that the azaindole core (\*9) (Figure 6) with a 2,4-substituted pyridine shown below was the minimum pharmacophore needed for a high binding affinity to NFTs. The insertion of fluorine in the minimal pharmacophore led to a loss in affinity by >10 fold. This was also observed when either pyridinyl rings were fluorinated. A phenomenon, which hinted at a specific electronic contribution of the basic nitrogen to NFT binding. In [18F]MK-6240 there is a minimum effect of fluorine on the basicity of the heterocyclic nitrogen in the isoquinoline ring and the presence of a primary amine, an additional stronger basic center, must have improved its affinity to NFTs, Kd 0.36, in comparison to the 1,6-naphthyridine derivative (\*12), with both basic centers in the ring, with affinity to NFTs, Kd 52.6 [202] (Table 3)

It exhibited favorable pharmacokinetics, with a fast brain uptake and clearance (Table 3). Uptake was higher in AD subjects and was considerably higher in brain regions expected to have NFT like in the hippocampus, but very low uptake in the cerebellar gray matter suggests a potential use of the cerebellar gray matter as a reference region. Based on reliability analysis simplified quantitative approaches could offer informed estimates of NFT load [203]. Furthermore, the spatial patterns of binding of the tracer were in accordance with the neuropathological staging of NFT, as reported from recent clinical studies [204].

Preclinical findings confirmed a lack of binding to MAO-A and MAO-B [203]. Unlike flortaucipir and [18F]THK-5351 off-target binding was not seen in the choroid plexus and basal ganglia [204], but like flortaucipir and various tau PET tracers, off-target binding to neuromelanin- and melanin containing cells like the pigmented neurons in the substantia nigra, and meninges was observed [52,187,201]. To confirm initial observations, there are ongoing clinical trials on non-AD patients. The phase I of its clinical trial was completed in 2016 [205].

A 1,5-napthyridine derivative, [18F]JNJ64349311(JNJ311) (Figure 6) was also reported. It showed moderate initial brain uptake but a fast brain clearance in biodistribution assay using wild-type mice (Table 3). It is quickly metabolized as was seen in NMRI mice and a rhesus monkey 30 min after intravenous injection, where only 22% and 35% respectively of the recovered activity was the intact radioligand. However, all the detected radiometabolites were more polar than the tracer, and none was found in the brain even at 60 min after injection. Furthermore, no bone uptake was detected in a rhesus monkey during a 120 min scan, in the duration of a microPET scan, which also showed a moderate initial brain uptake (SUV of 1.9 at 1 min p.i.) with a rapid wash-out.


**Table3.**BindingaffinitiesandPharmacokineticsoffeaturedtauPET-tracers.

 log partition (octanol/water) log partition (octanol/PBS) reported respective publications. Selectivity Aβ: Ki(Aβ)/Ki(tau).% inhibition of 10 nM of [3H]T808 on fresh frozen human brain sections derived from AD cases. 3 Self-competition. 4 In competition with [3H]PiB. 5 Peak uptake (injected dose pergram brain; ID/g); 6 ratio of peak uptake divided by peak at 30 min. 7 In competition with [18F]RO-948. 8 [ 3H]AV680 as competitor. 9 [ 3H]Florbetapir as competitor. 10 Data are expressed as SUV mean. NA: data not available. \*numbers given to the tracers in the respective publications.

#### *Molecules* **2020**, *25*, 977

2

Semi-quantitative autoradiography studies on post-mortem tissue sections of human AD brains displayed highly displaceable binding to NFT-rich regions, but it showed no specific binding to human PSP and CBD brain slices. Based on its in vitro and in vivo preclinical profiling, it was deemed a promising candidate for quantitative tau PET imaging in AD [206]. There is presently no in vivo human data for the tracers of the JNJ series.

#### *2.3. Selective PET-Tracers for the Imaging of* α*-syn*

Most of the PET/SPECT tracers developed and approved so far for the differential differential diagnosis of PD have been geared towards the evaluation of the function of the dopaminergic system [207]. As was mentioned earlier, 50% of substantia nigra cells (stage III of the Braak staging) and a probable loss of a higher percentage of dopaminergic nerve endings in the putamen have to be lost before the appearance of motor symptoms [71–73], in contrast, based on the findings of Braak et al., there is deposition of α-syn in LBs and LNs, which occurs sequentially and additively throughout the VI stages of disease progression [74], therefore the most accurate and earliest detection of premotor PD should be based on imaging α-syn instead of dopaminergic changes [208].

The development of α-syn PET tracers is still an unmet need and in its early stages. Regardless, efforts have been made in the past decades and are still being presently made to develop tracers with a high affinity and selectivity for α-syn over Aβ and tau aggregates.

#### 2.3.1. The Phenothiazine Derivatives

In 2011, to discover selective α-syn tracers Yu et al. synthesized a series of phenothiazine derivatives. Three of the tricyclic compounds (Figure 7) based on in vitro Th-T competition assay to recombinant α-syn fibrils were selected: [11C]SIL5, [125I]SIL23, and [18F]SIL26 based on the fact that they displayed an affinity (Ki) to the α-syn fibrils less than 60 nM [209] (Table 4).

In further tests, [125I]SIL23 with a ki of 57.9 nM [209] was able to bind to α-syn fibrils in postmortem PD brain homogenates, which indicated that the tracer binding affinity in PD brain samples is comparable to its affinity to recombinant α-syn fibrils. It also displayed 5-fold and 2-fold less affinity for Aβ1-42 and tau aggregates respectively in comparison to α-syn fibrils, however this selectivity was insufficient for in vivo imaging. Moreover, a high nonspecific binding in white matter seems to limit autoradiography with the tracer initial experiments. Furthermore, its affinity for α-syn fibrils Kd 148 nM, is also not optimal for the in vivo imaging of α-syn fibrils [210].

[ 11C]SIL5 and [18F]SIL26 with binding affinities (Ki) 1.8<sup>×</sup> and 1.2<sup>×</sup> more than that of [ 125I]SIL23 respectively, showed a low initial brain uptake in healthy Sprague-Dawley rats 5 min p.i. 0.953%ID/g and 0.758%ID/g respectively, and slow washout with [18F]SIL26 performing poorer of the 2 tracers with not less than 50% of the initial brain concentration at 5 min remaining at 60 min. [ 11C]SIL5 still had at 60 min 16.57% of its initial brain uptake at 2 min, which suggested a slow washout of the tracer or a lot of unspecific binding (Table 4).

In vivo microPET imaging in a healthy cynomolgus macaque confirmed that [11C]SIL5, with a faster washout kinetics of the two was able to penetrate the BBB into the brain, and also has a homogeneous distribution and fast washout kinetics. The authors believe that both compounds require further structural optimization in order to make them a more suitable α-syn tracer [211].

**Figure 7.** Structures of selective α-syn-PET tracers: [11C]SIL5, [125I]SIL23, [18F]SIL26, \*6a, \*14, \*20, \*46a, \*11a, \*11b, [125I]IDP-4, [18F]FS3-1 ([18F]DABTA-11). \*numbers given to the tracers in their respective publications.

2.3.2. The Indolinone and Indolinonediene Derivatives

An SAR study by structural modifications of an indolinone derivative \*6a [212] (Figure 7), by the introduction of different alkyl and arylalkyl groups substituted at the indolinone nitrogen led to the identification of an aza-analog-\*14 (Figure 7), which was the most successful in this series (Table 3). With a binding affinity Ki 79 nM to recombinant α-syn fibrils less than that of the select 3 phenothiazine tracers [209,213], with 1,4- and 11-fold selectivity over recombinant Aβ and tau aggregates. Another major limitation of the series was the presence of E/Z isomers, which re-equilibrate after separation.

A homologation of the double bond in \*6a by the addition of an additional double (to form diene group) bond, in order to increase affinity to α-syn fibrils over the other aggregates gave a series of compounds, which were a mixture of stereoisomers with either an E,E or Z,E configuration, which quickly re-equilibrated after chromatographic purification. \*20 (Figure 7) was however selected from the series based on its improved affinity (Ki) for recombinant α-syn fibrils 1.9× more than \*14, but unfortunately affinity Ki 27.6 nM for Aβ increased as well, with 1.3-fold selectivity over tau aggregates (Table 4). However, its strong fluorescent attributes allowed for a performance of fluorescent microscopy studies of postmortem AD and PD brain samples. Results showed that it labels both LB

and Aβ plaques. Regardless, this showed that indolinonediene derivatives can label α-syn fibrils in LBs.

Introduction of a para nitro group into the pendant benzene ring of the diene moiety made it possible to isolate both the *E,E* and *Z,E* stereoisomers, and further explore their in vitro properties. The *Z,E* regioisomers were generally more active than the corresponding *E,E* configuration in terms of higher affinity for α-syn over Aβ and tau fibrils. The best in this series was [18F]46a with the highest affinity (Ki) for α-syn fibrils 2 nM, 70-fold and 40-fold less affinity for Aβ and tau fibrils respectively (Table 4).

Unfortunately, due to its high lipophilicity log D 4.18, which prevented obtaining a reliable and reproducible results from binding assays to insoluble α-syn acquired from PD brain and the possible reduction of the nitro group to an amino group in vivo makes it an unsuitable PET probe for imaging LB and LN in PD subjects. In spite of the shortcomings of the compound it showed interesting selectivity for α-syn fibrils over Aβ and tau, and could serve as a good lead for further development of α-syn fibril tracers [213].

#### 2.3.3. Chalcone Derivatives and Structural Cogeners

Hsieh et al. went further to investigate a series of chalcone derivatives, whose enone moiety serves as an isosteric replacement of the diene group in the indolinonediene derivatives while precluding the *E,E* and *Z,E* isomerization problem. The indole ring was further replaced with a benzothiazole ring system, based on a previous SAR study, which revealed that an electron-deficient ring like the aza-indole system has a higher affinity for α-syn fibrils relative to the indole ring system and to prevent the Michael-acceptor properties associated with the chalcone system they replaced the enone moiety with an isoxazole and a pyrazole ring system. The results of a competition in vitro binding studies with Th-T led to the identification of a compound \*11a,b (37 a,b) (Figure 7), isoxazole derivatives with a modest affinity in comparison to [18F]46a (36) for α-syn at Ki 18.5 nM over Aβ and tau fibrils with 5-fold and over 54-fold less affinity respectively (Table 4). Although the compounds described in their report have modest affinity to serve as a PET radiotracer for in vivo imaging studies, they could, however could be used for further SAR studies [214].

Based on previous research, it was discovered that flavonoids could inhibit not only the formation of Aβ [215–217] but also of α-syn [218–220] aggregates, an indication that they could also bind with α-syn aggregates. With this in mind, Ono et al. developed some α-syn imaging tracers based on the chalcone scaffold: they developed four prospective chalcone derivatives (IDP compounds) with varying molecular lengths made possible by conjugated double bonds. A longer molecular length and a long conjugated π system were believed to lead to increased affinity of probes to Aβ [221], and tau aggregates as was seen in the PBB compounds [172,222].

All the IDP-compounds, unfortunately, displayed almost as much affinity for Aβ as they displayed for α-syn (Table 4), with affinity for α-syn increasing proportionately to molecular length, with not really much change in the selectivity to Aβ, which means that [125I]IDP-4 with a tetraene structure was the best in this regard. It had a high binding affinity to α-syn Kd of 5.4 nM and 3-fold selectivity to recombinant Aβ fibrils, which nevertheless was not as high as [18F]46a, which the authors believe could be attributed to different binding assay conditions.

[ 125I]IDP-4 (Figure 7) in vivo biodistribution in normal mice performed poorer than the other IDP-compounds, with a brain uptake of 0.45%ID/g at 2 min p.i. and a low brain clearance, 93.3% of the concentration at 2 min still remaining at 60 min (Table 4). The suboptimal pharmacokinetics could be due to the lipophilicity, high molecular weight and chemical structure of the tracer. In any case, this property makes it an unlikely tracer for in vivo imaging of α-syn aggregates. Still, it could be a useful probe for in vitro screening of compounds for affinity to α-syn and Aβ fibrils in vitro [222].


*Molecules* **2020** , *25*, 977

#### 2.3.4. Diarybisthiazole Compounds

More recently, highly innovative small molecules utilized as diagnostic probes, based on 4,4'-diaryl-2,2'-bithiazole, called DABTAs were developed by Yousefi et al. as sensitive, selective and specific tracers. According to pilot data these compounds suitable for the visualization and quantification of α-syn pathology by nuclear medicine imaging. With the help of these newly developed ligands, the presence, distribution and progression pattern of α-syn can be investigated in the living brain, reflecting the disease entity and stage. Preliminary results with the [18F]FS3-1 ([18F]DABTA-11) (Figure 7) in a rat model overexpressing human E46K mutated α-syn are promising regarding sensitivity [223,224]. In addition, in vivo imaging of the progression and spreading of α-synuclein pathology over time with the tracer greatly motivates the authors to develop tracers for aggregated α-syn in PD, DLB or MSA.

#### **3. Conclusions**

Several Aβ PET tracers have been developed and promoted discussion on their clinical values. Among these three tracers have been already approved by the FDA and EMA. Research on tau PET yielded several tau-tracers already entered into clinical investigations, however, still inherent are challenges in selective tau imaging. The lack of any consented and approved tracer for aggregated α-syn greatly motivated scientists in the field to develop tracer for aggregated α-syn in PD, DLB or MSA. Remarkable progress has been made so far in order to fulfill the unmet need for α-syn PET tracers with suitable pharmacokinetics, binding affinity and selectivity. Lately, innovative small molecules have been identified with excellent binding affinity and selectivity for α-syn fibrils relative to Aβ and tau aggregates. Utilizing high-throughput binding assays, in-silico design in the conventional tracer development may accelerate the identification of new leads for a specific α-syn PET imaging.

**Funding:** The APC was funded by Technical University of Munich, University Library Open Access Publishing Fund. BCU gratefully acknowledge the funding received towards his PhD from the German Academic Exchange Service (DAAD).

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Metabolic Brain Network Analysis of FDG-PET in Alzheimer's Disease Using Kernel-Based Persistent Features**

**Liqun Kuang 1,\*, Deyu Zhao 1, Jiacheng Xing 2, Zhongyu Chen 3, Fengguang Xiong <sup>1</sup> and Xie Han 1,\***


Academic Editor: Peter Brust

Received: 7 May 2019; Accepted: 20 June 2019; Published: 21 June 2019

**Abstract:** Recent research of persistent homology in algebraic topology has shown that the altered network organization of human brain provides a promising indicator of many neuropsychiatric disorders and neurodegenerative diseases. However, the current slope-based approach may not accurately characterize changes of persistent features over graph filtration because such curves are not strictly linear. Moreover, our previous integrated persistent feature (IPF) works well on an rs-fMRI cohort while it has not yet been studied on metabolic brain networks. To address these issues, we propose a novel univariate network measurement, kernel-based IPF (KBI), based on the prior IPF, to quantify the difference between IPF curves. In our experiments, we apply the KBI index to study fluorodeoxyglucose positron emission tomography (FDG-PET) imaging data from 140 subjects with Alzheimer's disease (AD), 280 subjects with mild cognitive impairment (MCI), and 280 healthy normal controls (NC). The results show the disruption of network integration in the progress of AD. Compared to previous persistent homology-based measures, as well as other standard graph-based measures that characterize small-world organization and modular structure, our proposed network index KBI possesses more significant group difference and better classification performance, suggesting that it may be used as an effective preclinical AD imaging biomarker.

**Keywords:** Alzheimer's disease (AD); network measure; graph theory; brain network; positron emission tomography (PET); persistent homology

#### **1. Introduction**

Alzheimer's disease (AD) is one of the most common neurodegenerative neurological diseases and is the most common form of dementia in the elderly [1]. Its clinical manifestations include long-term memory loss, cognitive decline, language disorders, and other symptoms. AD seriously affects the normal life of the elderly. However, the pathology of AD is not yet clear [1]. Some existing imaging technologies are used to explore the mechanisms of human brain function. Compared to magnetic resonance imaging (MRI), fluorodeoxyglucose positron emission tomography (FDG-PET) has been demonstrated to be a more precise predictor of both AD and mild cognitive impairment (MCI), and is more suitable for monitoring disease progression [2]. It collects and measures changes in glucose metabolism values in brain regions or local brain cells. The signals are then converted into effective three-dimensional images and the connectivity between brain regions are analyzed.

The topological organization of metabolic brain networks have been successfully characterized in many cases using various measures based on graph theory [3–5], such as characteristic path length (CPL) [6], global efficiency [7], modularity (Mod) [7,8], and network diameter (ND) [9], to name but a few. Specifically, in patients with AD and MCI, several research groups have reported topological alterations in the whole-brain connectome, including a loss of small-worldness [10], a redistribution of hubs [9,11], and a disrupted modular organization [12]. Traditionally, weighted networks usually require defining a set of thresholding values before quantifying network topology [13–15], which may result in inconsistent network features when the thresholding values vary. Generally, the choice of threshold is rather arbitrary and there are no widely accepted criteria [16,17].

Recently, persistent homology [18] in algebraic topology has been studied to detect persistent structures generated over all possible thresholds [19–23] in brain network analysis. There have been significant efforts to model evolution of brain networks and to link network topology to network dynamics. This method constructs multiscale network for all possible thresholds wherever the persistent topological features over the evolution of the network changes are identified. Its ability to handle noisy data and provide homological information has turned it into a successful tool for the analysis of brain network structures [24]. One typical application of persistent homology is in a Betti number plot (BNP) [18], which has been successfully applied to the brain network research on epilepsy [20], autism spectrum disorder, and attention-deficit hyperactivity disorder [19,23]. As BNP ignores the association between persistent features and forthcoming thresholding value changes, we proposed an integrated persistent feature (IPF) by integrating an additional feature of connected component aggregation cost with BNP, and applied it to measure an AD network using resting state functional MRI (rs-fMRI) in our prior study [25]. However, both BNP and IPF applied linear regression analysis for computing the slope of the plot over all thresholds as a univariate network index. Such a slope-based approach may not accurately characterize the changes of persistent features over graph filtration because the curves are not strictly linear. Moreover, our previous IPF works well on an rs-fMRI cohort though it has not been used to study metabolic brain networks yet.

In this paper, we borrow the idea of kernel methods [26,27] on persistent homology and propose a kernel-based IPF (KBI) index based on our prior work on IPF. We hypothesized that our KBI index may help to better reveal the difference between brain networks. With the cross-sectional FDG-PET imaging data of 140 AD, 280 MCI, and 280 normal control (NC) individuals, we set out to test this hypothesis by computing the KBI indices that measure the differences between AD, MCI, and NC groups. We further perform statistical inference and classification to validate the power of KBI.

#### **2. Results**

In this section, we use FDG-PET data to evaluate statistical power and classification performance of our proposed KBI index for the analysis of brain metabolic networks related to AD. We further compared it with prior persistent features, BNP [19,21,23] and SIP [25], as well as some other standard graph-based indices.

#### *2.1. Metabolic Brain Networks*

After data preprocessing, the summarized point cloud were extracted from PET 3D imaging using predefined automated anatomical labeling atlas with 90 (AAL-90) regions of interests (ROI) [28]. We obtained the SUV matrix for all 700 subjects in all 90 ROIs and plot three histograms, in Figure 1, to show the global distributions of FDG uptake in AD, MCI, and NC. As the number of AD is half of MCI and NC, we have normalized the SUV distribution of AD by doubling its statistics. We observed that the AD cohort has lower glucose metabolism than MCI and NC, but no significant differences were detected in the statistical inference of permutation test. We calculated the Pearson-based correlation distance of FDG uptake between each pair of brain regions using Equation (1) and constructed group-wise brain metabolic networks. The three multiscale networks of AD, MCI, and NC groups are shown in Figure 2, which visualizes the evolution of brain networks over different thresholds.

**Figure 1.** The fluorodeoxyglucose (FDG) uptake distribution of AD, MCI, and NC groups.

**Figure 2.** The constructed multiscale networks of AD, MCI, and NC by graph filtration λ, and the node color represents the ROI index predefined in AAL-90 atlas.

#### *2.2. Brain Network Features*

We computed the values of graph-based network indices (CPL, ND, and Mod) in three groups based on their weighted networks after filtering their edges, whose corresponding *p*-values passed a statistical threshold (Bonferroni corrected *p* < 0.05). We then obtained the multiscale network according to graph filtration and computed the BNP and IPF index (i.e., SIP), as well as KBI index. Figure 3 shows three separate IPF plots of AD, MCI, and NC. All brain network index values are shown in Table 1. The differences between groups need to be further verified by statistical inference and classification.

**Figure 3.** The proposed integrated persistent feature (IPF) plot for three group-wise networks of AD, MCI and NC, respectively.



#### *2.3. Statistical Group Di*ff*erence Performance*

In this study, we use the permutation test for 10,000 permutations between any two groups, and show the resulting *p*-value in Table 2. Only the proposed KBI index obtained a significant difference in any between-group at the significance level of 0.05.

**Table 2.** Statistical *p*-values for between-group differences by different graph indices on an AAL-90 atlas.


Only the proposed KBI index detected significant difference in any between-group (any *p*-value < 0.05).

#### *2.4. Classification Performance*

Furthermore, we resampled the networks 5000 times for each group with the resampling rate of 0.5, and obtained 5000 values of each network index for each group. We then performed leave-one-out crossvalidation to evaluate the classification powers of two-label (Figures 4 and 5) and three-label (Figure 6) by SVM. Our KBI shows better classification performance than prior persistent features, SIP and BNP, as well as other standard graph-based features, including CPL, ND, and Mod.

**Figure 4.** Comparison of ROC curves of different network indices for MCI vs. NC.

**Figure 5.** Comparisons of ROC curves of different network indices for AD vs. NC.

**Figure 6.** Three-label classification.

#### **3. Discussion**

#### *3.1. Present Findings*

This study has three main findings.

First, from Figure 1, we found that the AD cohort has lower glucose metabolism than MCI and NC. This may imply cognitive impairment in AD and MCI. Such an inference is further partly confirmed by graph theory analysis because a larger CPL is present in AD and MCI, while the network with smaller CPL is considered to be efficient.

Second, in our previous study [25], we had developed a univariate network index, SIP, based on homology to model graph dynamics over all possible scales and applied it to study the rs-fMRI data of AD. We found the SIP values of AD were lower than MCI and much lower than NC. In the current PET data, we still find the SIP values show the same pattern AD < MCI < NC, suggesting a slower network integration rate in AD and MCI groups. Thus, the results from both independent cohorts provide consistent empirical evidence for decreased functional integration in AD dementia and MCI.

Finally, we propose a novel univariate network index KBI to enhance our previous study based on persistent homology. With our univariate KBI index, the difference of persistent features between cognitive dysfunction and NC brain network can be measured more accurately. Our preliminary experimental results demonstrate that the proposed KBI may greatly boost prior SIP and BNP power in both statistical inference and classification analyses. The KBI also outperforms other standard graph-based methods, such as CPL, ND, and Mod, suggesting that our method may serve as a valuable preclinical AD imaging biomarker.

#### *3.2. Exploring Other Connectivity Definitions*

There are many types of distance functions to construct weighted networks in brain network analysis [29], such as Pearson correlation, partial correlation, psycho–physiological interactions, ReHo, partial least squares, wavelet-based correlation, mutual information, synchronization likelihood, principal component analysis, independent component analysis, cluster analysis, dynamic causal modeling, Granger causality modeling, structural equation modeling, and multivariate autoregressive modeling, among others. At present, it is difficult to put forward the evaluation criteria of these methods, and few studies have compared them comprehensively. Although the Pearson correlation that we used in this study may be the most practical scheme to define the connectome in AD studies, there is still debate about the choice of connectivity definition [29,30]. Therefore, we performed four other connectivity definitions to explore more potentials in defining connectivity network. They were Kendall correlation [31], Spearman correlation [32], partial least squares [33], and Granger causality modeling [34]. The obtained *p*-values of our proposed KBI with these distance functions are shown in Table 3. There was no significant difference if Granger causality modeling was used, while the other three methods detected at least one significant difference. It should be noted that none of these methods performed significantly better than Pearson correlation (Table 1) in our current dataset. Moreover, when we checked all measures to discriminate AD, MCI, and NC by these methods, we found that the three-label classification accuracy of BNP (88.3%) was improved greatly if the partial least squares method was applied, while the performances in other cases have not been improved significantly. Such an empirical study may justify the connectivity definition adopted in our current work.


**Table 3.** Statistical *p*-values for between-group differences of KBI by different connectivity definitions.

#### *3.3. Ways of Network Construction*

Graph-based brain connectome analyses are sensitive to the choice of parcellation schemes. To assess the effects of different parcellation strategies, we carried out the same set of analyses with another commonly employed atlas, the Harvard–Oxford atlas [35,36] with 110 ROIs (HOA-110). The detailed statistical significances of between-group difference on HOA-110 are presented in Table 4. Again, our proposed KBI achieved better statistical power.


**Table 4.** Statistical *p*-values for between-group difference of different network indices on HOA-110 atlas.

Similarly, only the proposed KBI index detected significant difference in any between-group.

In the metabolic network construction, the common practice is building a group-wise brain network for each group as there is only one summarized value (average SUV) in each ROI. However, we notice that some studies [37] constructed subject-wise networks by dividing each ROI of a subject into blocks to obtain the correlation distance between any two ROI. Thus, subject-wise networks were constructed. We did not study this method in as it would require additional discussion that would exceed the scope of this paper.

In addition, we defined the connectivity between two brain regions as 1-Pearson correlation in Equation (1) in our study. Although some existing studies [19,23] also define such kinds of connectivity in analyzing brain network properties, the common practice in brain network analysis based on graph theory is to directly specify the Pearson correlation as the edge weight. To assess the effect of different connectivity definitions on other graph-based measures, we performed statistical inference on the brain networks whose edges were defined as directly based on Pearson correlation, and the statistical *p*-values of graph-based measures are shown in Table 5. We found the results were different from the previous results in Table 2, suggesting that graph-based measures could be affected by way of connectivity definition. We also found that none of the graph-based measures could detect all between-group differences significantly in either connectivity definition.

**Table 5.** Statistical *p*-values for between-group differences by specifying Pearson correlation as connectivity directly.


#### *3.4. Limitations and Future Work*

Despite the promising results obtained by applying our proposed network index KBI based on persistent homology to PET, there are three important caveats. First, the current study only takes the zeroth persistent homology into account. Higher-order persistent features are also worth studying. In future, we will try to improve the performance of our method by considering higher dimensional persistent homology, such as the first Betti number, which is designed to calculate the number of holes in a graph and may boost the performance, especially in sparse networks that tend to have more holes. Then, although the subject-wise network is more convenient, efficient, and useful than group-wise network for brain network analysis, as we discussed in our prior study [25], we only measured the group-wise metabolic brain network according to regular practice in PET data analysis. In future, we will validate the KBI in a subject-wise metabolic network. Finally, this study was based on cross-sectional PET analysis, and we compared their network indices. With longitudinal PET analysis, we may further study the evolution between longitudinal brain networks by quantifying the difference of their persistent features.

#### **4. Materials and Methods**

Figure 7 shows the pipeline of our framework, where the data flow from FDG-PET brain images to some network indices. The details are described in following subsections.

#### *4.1. Participants*

Data used in the preparation of this article were obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu) [38,39]. ADNI was launched in 2003 as a public–private partnership led by Principal Investigator Michael W. Weiner, MD. The primary goal of ADNI has been to test whether serial magnetic resonance imaging (MRI), positron emission tomography (PET), other biological markers, and clinical and neuropsychological assessment can be combined to measure the progression of mild cognitive impairment (MCI) and early Alzheimer's disease (AD).

In this study, we chose 700 subjects with FDG-PET data from ADNI2. To match the three research cohorts of AD, MCI, and NC in gender and age, 140 AD, 280 MCI, and 280 NC subjects from 57 sites across North America were selected. The detailed cohort information is described in Table 6.

**Figure 7.** The pipeline of the brain network analysis based on a group of subjects. SUV, support vector machine; BNP, Betti number plot; SIP, slope of IPF plot; KBI, kernel-based IPF; CPL, characteristic path length; ND, network diameter; Mod, modularity.

**Table 6.** Demographic information of the subjects in this study.


Key: AD, Alzheimer's disease; MCI, mild cognitive impairment; NC, normal control; <sup>a</sup> mean ± SD; <sup>b</sup> male/female number; <sup>c</sup> statistical group significance using ANOVA test.

#### *4.2. FDG-PET Data Acquisition and Preprocessing*

All FDG-PET scans were obtained using Siemens, GE, and Philips PET scanners. Details of the PET data acquisition is described at http://adni.loni.usc.edu/methods/pet-analysis/pre-processing/. All FDG-PET scans used in this research are preprocessed (step 1 of Figure 7) as follows [40]. First, in order to eliminate the individual differences in brain morphology between subjects such that they can completely coincide and be subject to effective statistical analysis, we used the software toolkit Statistical Parametric Mapping (SPM8) [41] in MATLAB (Mathworks Inc, Natick, MA, USA) to linearly align the images into the Montreal Neurological Institute (MNI) space using the TPM.nii template file released with SPM. Second, we borrow a brain mask from SPM, exclude the brain stem and only keep the cerebral cortex (because the cerebral cortex is the object of this study), and then segmented all the images with this cerebral mask. Third, we conducted spatial smoothing with a Gaussian kernel of the full width at half maximum (FWHM) equal to (8,8,8) in three directions (x,y,z) to improve signal-to-noise.

#### *4.3. Network Construction*

A weighted graph is a natural and efficient way to represent metabolic brain network because it represents a discretized version of original PET images. In computer graphics, polygon meshes, as a class of graphs with particular connectivity restrictions, are extensively used to represent the topology of an object [42]; however, the mesh representation may not be the most suitable representation for analyzing PET images because of connectivity restrictions [43]. Here, we extend polygon meshes to general graphs by relaxing the connectivity restrictions. Such graphs are easier to construct and are flexible enough to capture metabolic information. We construct a weighted network by encoding the metabolic information through an adjacency matrix *W* = {*wij*}. The node corresponds to the brain regions, and the edge corresponds to the interregional correlation of brain metabolism. Specifically, the region parcellation in brain imaging is usually defined based on an anatomical atlas. In this study, we applied a predefined atlas, an automated anatomical labeling atlas with 90 (AAL-90) regions of interests (ROI) [28]. Once an ROI is specified, an overall summary measure within it can be calculated to assess the response as a whole, rather than on a voxel-by-voxel basis (step 2 of Figure 7). The most straightforward way to do so is by taking the average standard uptake values (SUV) of all voxels within the ROI. The SUV of a specific ROI is <sup>1</sup> *M* - *M p* = 1 *vp*, where *M* is the total voxel number

in a given ROI and *vp* is the FDG uptake value of voxel *p*. Given *K* subjects and *N* brain regions, let *SUVi* = {*SUVi*1, *SUVi*2, ... , *SUViN*,}(1 ≤ *i* ≤ *N*) be the vector of average SUV in *i*-th ROI of all *K* subjects (step 3 of Figure 7), and the edge weight *wij* between two brain regions is defined as 1-Pearson correlation of SUV between them (step 4 of Figure 7), i.e.

$$w\_{ij} = 1 - \frac{cov\{SLIV\_{i\prime}SLV\_j\}}{\sigma\_{SLV\_i}\sigma\_{SLV\_j}},\tag{1}$$

where *SUVi*, *SUVj* are the average SUV in *i*-th and *j*-th brain region respectively, *cov* is the covariance, <sup>σ</sup> is the standard deviation, and *cov*(*SUVi*,*SUVj*) <sup>σ</sup>*SUVi* <sup>σ</sup>*SUVj* is coefficient of Pearson correlation.

#### *4.4. Network Indices*

In clinical settings, doctors prefer single indices as biomarkers because a single neuroimaging index provides a practical reference for evaluating disease progression and for effective treatments. Generally, there are some available network indices based on graph theory that measure brain global attributes. In addition, we focus on some univariate indices that were developed from persistent homology in algebraic topology, and compare them with the network indices from traditional graph theory in our experiments.

#### 4.4.1. Traditional Graph Theory Indices

Traditionally, graph theoretical analysis has been applied to measure brain network topological features. In this study, three global network indices based on graph theory are investigated, including characteristic path length (CPL) [6], network diameter (ND) [9], and modularity (Mod) [7].

Briefly, CPL can be understood as indicating a network with "easily" transferred information. It is the average shortest path length between all pairs of nodes in the graph, and is calculated as *CPL* = <sup>1</sup> *N*(*N*−1) - *i*∈*V*,*j*∈*V*,*ij di*,*j*, where *di*,*<sup>j</sup>* is the shortest path length between nodes *i* and *j*. Note that infinitely long paths (i.e., paths between disconnected nodes) are not included in computations. ND is the greatest distance between any pair of nodes, and is defined as *ND* = *max <sup>i</sup>*∈*<sup>V</sup> max <sup>j</sup>*∈*<sup>V</sup> di*,*j*. It enables understanding of the size of a network. A graph with a large ND and small CPL would therefore be considered an efficient network. Mod describes the extent to which a network has modules that differ from others, each of which is independent and functionally specialized [7]. Computationally, it is ⎡ ⎛ ⎞ 2⎤

expressed as *Mod* = *i*∈*M* ⎢⎢⎢⎢⎢⎣ *cii* − ⎜⎜⎜⎜⎝ *j*∈*M cij* ⎟⎟⎟⎟⎠ ⎥⎥⎥⎥⎥⎦ , where *i* and *j* are individual modules in the set of all modules *M*, and *c* is the proportion of existing connections between two modules.

In practice, we filtered the weighted network before computing these graph-based indices by only selecting the edges whose corresponding *p*-values passed through a statistical threshold (Bonferroni corrected *p* < 0.05) and then adopted the Brain Connectivity Toolbox (https://sites.google.com/site/ bctnet/) [7] for their implementation (step 6 (right) of Figure 7).

#### 4.4.2. Persistent Features Based on Persistent Homology

Persistent homology is an emerging mathematical concept for characterizing shapes in complex data, and the persistence features based on BNP are widely recognized as a useful feature descriptor. BNPs can distinguish robust and noisy topological properties over a wide range of graph filtrations based on the connectivity of *k*-dimensional simplicial complexes [18] (step 6 (left) of Figure 7). Graph filtration is an important tool [24] in persistent homology that constructs nesting subnetworks in a coherent manner and avoids thresholding selection (step 5 of Figure 7). BNPs have been successfully applied to measure brain networks based on FDG-PET and structural MRI data [23] in some neurodegenerative diseases. In our previous study [25], we proposed an integrated persistent feature (IPF) by integrating an additional feature of connected component aggregation cost with BNP to achieve holistic descriptions of graph evolutions. The IPF at filtration λ*<sup>i</sup>* is defined as [25]

$$IPF\_{\lambda\_i} = \begin{cases} \frac{m-i}{m(m-1)} \sum\_{k=i+1}^{m-1} \lambda\_{k'} & 0 \le i \le m-2 \\\\ & \text{ } \end{cases} \tag{2}$$

where the maximal graph filtration is λ<sup>0</sup> = 0 < λ<sup>1</sup> < λ<sup>2</sup> < ··· < λ*m*−1. As the IPF plot over all possible filtration values is a monotonically decreasing convergence function, the absolute value of the slope of IPF plot (SIP) was defined as a univariate network index and was successfully applied to quantify brain network dynamics on rs-fMRI data of AD. Both the BNP and SIP indices indicate the rate of connecting components converging over the filtration value, and can be thought as the information diffusion rate or the convergence speed with said network.

#### 4.4.3. The Kernel-Based IPF (KBI) Index

Although the SIP has been developed as a univariate network index in our previous study, it may not be the most appropriate way to describe IPF plot as it is nonlinear, strictly speaking. Recently, some kernel methods [26,27] have been defined on persistent homology to measure the distance between persistence diagrams, which are not only provably stable but also discriminative. Therefore, we employ the framework of kernel embedding of the IPF plot into reproducing kernel Hilbert spaces [26]. Given a point set of IPF plot *X* = {*x*1, *x*2, ··· *xN*} and a template *T* = {*t*1, *t*2, ··· *tN*} that are obtained from an average metabolic network of all NC subjects, the kernel-based IPF (KBI) index is defined as

$$KBI(X) = \frac{1}{N} \sum\_{\substack{x\_i \in X, \ t\_i \in T, \ i = 1}}^{N} \tan^{-1} \left( \mathbb{C} \left( \boldsymbol{\lambda}\_i^X \right)^p \right) \tan^{-1} \left( \mathbb{C} \left( \boldsymbol{\lambda}\_i^T \right)^p \right) e^{-\frac{\frac{x\_i - t\_i}{2\sigma^2}}{2\sigma^2}} \tag{3}$$

where *tan*−1(*C* λ*X i* )*p* and *tan*−1(*C* λ*T i* )*p* are both increasing functions with respect to maximal graph filtrations λ(*X*) of *X* and λ(*T*) of *T*, and are used for weighting the persistence (λ is a sequence of persistence of zeroth homology in fact). Hence, an essential persistence gives a large weight and a noisy persistence produces a small weight. By adjusting the parameters *p*, σ, and *C*, we can control the effect of the persistence. In our practice, we set *p* = 5,

$$\sigma = \underset{\forall X}{\operatorname{median}} \{ \underset{\mathbf{x}\_i, \mathbf{x}\_j \in X, \ i < j}{\operatorname{median}} \| \mathbf{x}\_i - \mathbf{x}\_j \|\},\tag{4}$$

$$\mathbf{C} = \underset{\mathbf{V}\boldsymbol{\lambda}^{\mathbf{X}}}{\operatorname{median}} \underset{\boldsymbol{\lambda}\_i^{\mathbf{X}} \in \boldsymbol{\lambda}^{\mathbf{X}}}{\operatorname{median}}}{\operatorname{median}} \{\boldsymbol{\lambda}\_i^{\mathbf{X}}\})^{-p}{}\_{\boldsymbol{\lambda}^{\mathbf{X}}}\tag{5}$$

so that they take close values to many *xi* − *ti* and λ*<sup>X</sup> <sup>i</sup>* , respectively.

#### *4.5. Statistical Analysis*

We applied a group-wise statistical analysis of permutation test to all network indices in the last section between AD, MCI, and NC groups. As there are no prior studies on the statistical distribution of any network index, it is difficult to construct a parametric test procedure. Moreover, as there is only one group-wise network for each group, it is necessary to empirically construct the null distribution and determine the *p*-value. The steps of our permutation test are described as follows. First, the actual network index difference in means between two groups is calculated according to the actual grouping of their subjects. Second, the subjects are randomly assigned to two groups, each of which is assigned the same group size. We then construct two group-wise networks based on such permutated groups and recalculate their indices, whose difference is recorded. This process is repeated 10,000 times and 10,000 permuted differences are obtained. Finally, the total number of permuted differences larger than the actual difference is counted and divided by 10,000. The obtained value is the probability of no difference between the groups, that is, the *p*-value.

#### *4.6. Classification*

We evaluate the power of our method by classification analysis. In this study, our proposed KBI index and other comparison network indices have only one univariate feature to discriminate the global brain network structure. Since the samples are limited, we apply resampling technology beforehand. For a specific group, half subjects are removed randomly at a time and the remaining subjects are used to construct a group-wise network. Resampling all subjects for *n* times repeatedly in each group until the results are stable, we can obtain *n* group-wise networks. Each group-wise network is an average of all involved subjects graphs. In our practice, the number of resampling times is set to 5000 (*n* = 5000) and we obtain 5000 resampled networks for each group. Each network can yield a KBI index and other comparable network indices. Then, we compute the values of proposed KBI index, as well as other network measures for all resampled networks, and run support vector machine (SVM) [44] on them. We conduct leave-one-out crossvalidation experiments to evaluate the classification performance. The classification accuracy, sensitivity (i.e., true positive rate), specificity (i.e., true negative rate), and area under the curve (AUC) of receiver operating characteristic (ROC) are severed as criterion of classification performance.

#### **5. Conclusions**

This work proposed a novel network index KBI based on our prior work of persistent feature IPF to measure the metabolic brain network of FDG-PET on cognitively impaired cohort. The proposed KBI encoded a great deal of dynamic information over all possible scales that may be inaccessible by standard graph-based measures. Compared to previous slope-based approaches of persistent homology, our kernel-based network index is more accurate regarding the characterization of differences between persistent features. Our current results show that the slope of IPF plot present a pattern AD < MCI < NC in a FDG-PET cohort, consistent with our prior finding in an rs-fMRI cohort, and indicate a slower network integration rate in AD dementia and MCI. Moreover, the enhanced measurement KBI greatly boosted the performance of prior persistent features and outperformed some standard graph-based network indices in both statistical inference and classification experiments, suggesting that our method may serve as a valuable preclinical AD imaging biomarker.

**Author Contributions:** Conceptualization and methodology, L.K., D.Z. and X.H.; validation: D.Z., J.X. and Z.C.; formal analysis: L.K., F.X., and X.H.; writing: L.K., D.Z., and X.H. visualization, F.X.; software, J.X. and Z.C.

**Funding:** This work was supported by the National Natural Science Foundation of China (61672473 and 61602426 to L.K., F.X. and X.H.) and Shanxi Province Key R&D Technology Project (201803D121081 to L.K., F.X. and X.H.).

**Acknowledgments:** Data collection and sharing for this project was funded by the Alzheimer's Disease Neuroimaging Initiative (ADNI) (National Institutes of Health Grant U01 AG024904) and DOD ADNI (Department of Defense award number W81XWH-12-2-0012). ADNI is funded by the National Institute on Aging, the National Institute of Biomedical Imaging and Bioengineering, and through generous contributions from the following: AbbVie, Alzheimer's Association; Alzheimer's Drug Discovery Foundation; Araclon Biotech; BioClinica, Inc.; Biogen; Bristol-Myers Squibb Company; CereSpir, Inc.; Cogstate; Eisai Inc.; Elan Pharmaceuticals, Inc.; Eli Lilly and Company; EuroImmun; F. Hoffmann-La Roche Ltd. and its affiliated company Genentech, Inc.; Fujirebio; GE Healthcare; IXICO Ltd.; Janssen Alzheimer Immunotherapy Research & Development, LLC.; Johnson & Johnson Pharmaceutical Research & Development LLC.; Lumosity; Lundbeck; Merck & Co., Inc.; Meso Scale Diagnostics, LLC.; NeuroRx Research; Neurotrack Technologies; Novartis Pharmaceuticals Corporation; Pfizer Inc.; Piramal Imaging; Servier; Takeda Pharmaceutical Company; and Transition Therapeutics. The Canadian Institutes of Health Research is providing funds to support ADNI clinical sites in Canada. Private sector contributions are facilitated by the Foundation for the National Institutes of Health (www.fnih.org). The grantee organization is the Northern California Institute for Research and Education, and the study is coordinated by the Alzheimer's Therapeutic Research Institute at the University of Southern California. ADNI data are disseminated by the Laboratory for Neuro Imaging at the University of Southern California.

**Conflicts of Interest:** The authors declare no competing financial interest.

#### **References**


**Sample Availability:** Samples of the compounds are not available from the authors.

© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
