Considerations in the Development of Reversibly Binding PET Radioligands for Brain Imaging

Abstract

The development of reversibly binding radioligands for imaging brain proteins in vivo, such as enzymes, neurotransmitter transporters, receptors and ion channels, with positron emission tomography (PET) is keenly sought for biomedical studies of neuropsychiatric disorders and for drug discovery and development, but is recognized as being highly challenging at the medicinal chemistry level. This article aims to compile and discuss the main considerations to be taken into account by chemists embarking on programs of radioligand development for PET imaging of brain protein targets.

1. INTRODUCTION

A strong premise in modern medicine is that neurochemical dysfunctions underpin many neuropsychiatric disorders. Indeed, the majority of drugs that are available for treating such disorders exert their effects through binding to one or more proteins within the brain [1]. Targeted proteins are commonly neurotransmitter transporters, receptors, ion channels, or enzymes (Table 1). Few techniques are available for investigating brain disorders at the molecular level through measurement of specific protein-ligand or protein-drug interactions in vivo. Positron emission tomography (PET) and single photon emission computed tomography (SPECT) have unrivaled sensitivity for this purpose when used with biochemically-specific radioligands. Proteins present in brain at very low regional concentrations (~ nM) may be detected and measured, albeit with a spatial resolution that is limited to a few millimeters. An ability to make such measurements allows the roles of brain proteins in the unfolding of neuropsychiatric disorders to be investigated [2–4]. Moreover, the interactions of drugs or experimental therapeutics with brain proteins may be assessed. Therefore, these molecular imaging techniques have become integral to CNS drug discovery and development programs [5–8], and are also important for monitoring the neurochemical effects of drug treatments or therapies [8].

Table 1.

Examples of drugs for treating neuropsychiatric disorders, and their respective protein targets.

Drug	For treating:	Main putative site(s) of action	Type
Alprazolam	Anxiety, Panic disorder	GABAA Receptor (Bz binding site)	Channel
Amphetamine	ADHD, Narcolepsy	DAT	Transporter
Ariprazole	Psychosis, Depression	D2, 5-HT1A and other receptors	GPCRs
Buprenorphine	Opiate addiction, Pain	μ-Opiate receptor	GPCR
Duloxetine	Depression, Anxiety	SERT and NET	Transporters
Escitalopram	Depression, Anxiety	SERT	Transporter
Fluoxetine	Depression, Anxiety	SERT	Transporter
Galantamine	Alzheimer’s disease	AChE	Enzyme
Lorazepam	Anxiety, Panic disorder	GABAA receptor (Bz binding site)	Channel
Memantine	Alzheimer’s disease	NMDA Receptor	Channel
Methylphenidate	ADHD	DAT	Transporter
Pregabalin	Neuropathic pain, Anxiety	Ca2+ channel (α2δ subunit)	Channel
Quetiapine	Bipolar disorder, Schizophrenia	D2, 5-HT2A Receptors, and others	GPCRs
Rasagiline	Parkinson’s disease	MAO-B	Enzyme
Risperidone	Schizophrenia, Bipolar disorder	D2 and 5-HT2 receptors	GPCRs
Sertraline	Depression, Anxiety, OCD, PTSD, PMDD	SERT	Transporter
Trazodone	Depression, Anxiety	SERT	Transporter
Venlafaxine	Depression, Anxiety	SERT, NET and DAT	Transporters

Key to further exploiting the potential utility of SPECT and PET in neuropsychiatric research is the development of useful radioligands. Primarily, because of the requirement for facile entry from blood into brain (see Section 8), suitable radioligands are based on low molecular weight drug-like molecules. Radioligands for use with SPECT are mostly labeled with γ-emitting iodine-123 (t_1/2 = 13.2 h). Only rarely has technetium-99m (t_1/2 = 6.0 h) been applied with some success [9, 10]. Both iodine-123 and technetium-99m are available commercially. A radioligand for use with PET must be labeled with a positron-emitter, which is nearly always carbon-11 (t_1/2 = 20.4 min) or fluorine-18 (t_1/2 = 109.8 min). The short half-lives of these positron-emitters require that they are produced close to or at the site of use. Nowadays, dedicated biomedical cyclotrons are invariably used to produce these PET radionuclides, whereby they are obtained in very simple chemical forms, such as [¹¹C]carbon dioxide, [¹¹C]methane or [¹⁸F]fluoride ion. Ensuing transformations of these radionuclide sources into radioligands for human use is complex and resource-demanding. PET is therefore more expensive and more logistically demanding than SPECT. However, PET has no need for physical collimation in order to achieve optimal spatial resolution, and therefore has far greater sensitivity for radiation detection than SPECT (Fig. 1). Moreover, small drug-like molecules offer more opportunities for labeling with either carbon-11 or fluorine-18 than with any of the heavier single γ-emitter options. PET, because of these advantages, has now emerged as far preferable to SPECT for brain molecular imaging.

Fig. (1).

Physical basis of PET brain imaging. A PET radioligand that has entered brain after intravenous administration emits a positron. This positron loses kinetic energy over a short distance of a few millimeters before combining with an electron to form positronium, which then annihilates to produce two 511 keV γ-rays that are emitted in almost exactly opposite directions. The PET camera has several adjacent rings of sensitive γ-ray detectors that record all such coincident arrivals of paired γ-rays. Lines between pairs of simultaneously activated detectors cover the positions of positron annihilation events from which PET scans can be reconstructed to measure the distribution of radioactivity in three dimensions. Rapid sequences of scans are recorded to provide regional kinetic information over a few half-lives of the radioligand label (i.e.; ¹¹C or ¹⁸F). The detection of coincident γ-ray arrivals amounts to ‘electronic collimation’, and dispenses with the need for ‘physical collimation’ which would otherwise diminish detection sensitivity as in SPECT.

Expansion of the biomedical scope of PET is primarily determined by the range of effective radiotracers that can become available. The development of PET radioligands for imaging proteins in brain is an especially challenging area, because the design of any single radioligand needs to satisfy a wide array of chemical, biochemical and pharmacological requirements. Over recent decades, the assorted properties to be sought in candidate PET radioligands for brain imaging have been surveyed occasionally, but at varying levels of detail [11–19]. This article revisits this topic taking into account new developments and findings, and is mainly intended to be a useful primer for chemists entering this research area. The following discussion is confined to radioligands that bind reversibly to their protein targets because these radioligands represent the vast majority in use for brain protein quantification. Nonetheless, the reader will readily appreciate that many of the considerations that apply to reversibly-binding PET radioligands apply just as well to SPECT and irreversible radioligands.

2. PRINCIPLES UNDERLYING BRAIN PROTEIN MEASUREMENTS WITH REVERSIBLY BINDING PET RADIOLIGANDS

Fundamentally, PET has the capability to provide quantitative measures of radioactivity distribution in living subjects, ranging from small animals (e.g., mouse, rat, cat, pig and dog) through to non-human primates (NHPs) and humans. Researchers in neurology and psychiatry aim to exploit this capability with suitable radioligands to derive quantitative information on the distribution of proteins of interest in living brain, and on how these distributions are changed in neuropsychiatric disorders, or by attempts at their treatment. Furthermore, drug developers can use target-specific radioligands to measure the extent of engagement of the target by established therapeutics or experimental drugs and so arrive at safe and effective dosage regimens in clinical practice or preceding clinical trials.

Before starting to discuss general considerations with regard to PET radioligand development, it is pertinent to provide a simple background on how reversibly binding radioligands serve to quantify the density of a specific brain protein of interest, since in turn this will start to define the properties that will be desirable in such radioligands. In reality, the derivation of output measures of protein density from PET imaging with reversibly binding radioligands constitutes a challenging scientific task - and one that remains an active area of methodological research. Readers are therefore directed to expert reviews of this area for more details [20, 21]. Here, only some of the bare essentials of major approaches to quantification of protein density are outlined as an aid to understanding the remainder of this article.

High-affinity reversible binding of a radioligand (R*) to a target protein (P) is at the foundation of the majority of PET methods for quantifying brain proteins (Equation 1).

$$\mathrm{R}\ast +\mathrm{P}\underset{k_{\text{off}}}{\overset{k_{\text{on}}}{\rightleftharpoons }}\mathrm{R}\ast \mathrm{P}$$ Equation 1

The rate of protein-radioligand association, k_on is a second order rate constant having units of nM⁻¹min⁻¹, whereas the rate of dissociation k_off is a first order rate constant having units of min⁻¹. The equilibrium dissociation constant K_D is simply defined as k_off/k_on. Importantly, it should be noted that the K_D value of a ligand for binding to a particular protein actually corresponds to the concentration of radioligand needed to bind half the available concentration of protein. The inverse of K_D defines the strength of ligand binding, and in fact ligand affinity (1/K_D) relates to the Gibbs free energy of binding (ΔG^O) by Equation 2:

$$\mathit{\Delta }{G}^O=-RT\text{ln}{K}_{\mathrm{D}}$$ Equation 2

in which R is the gas constant and T is the absolute temperature. According to this relationship a radioligand with a K_D of 1 nM for a target protein has a binding energy of 53 kJ/mol, which is much lower than that of a typical covalent bond (e.g., ~ 430 kJ/mol for a C-H bond).

For brain protein imaging with PET to be successful, the concentration of bound radioligand ([R*P]) will clearly need to be high relative to unbound radioligand concentration [R*], which implies that numerically k_on should be high and k_off relatively low. Consequently, successful PET radioligands usually have affinities represented by K_D in the low nM or, in some cases, sub-nM range.

The concentration of the target protein, often called B_max, is a further vital consideration. Application of the Law of Mass Action to the simple theoretical equilibrium shown in Equation 1 can be used to derive Equation 3, which relates the concentration of bound protein B to both B_max and K_D.

$$B=([\mathrm{R}\ast ]\times B_{max})/(K_{\mathrm{D}}+[\mathrm{R}\ast ])$$ Equation 3

[R*] represents the concentration of free radioligand, which includes any non-radioactive counterpart, known as carrier. As [R*] increases, B asymptotically approaches B_max. However, under conditions where [R*] is much less than K_D, Equation 3 approximates to:

$$B=([\mathrm{R}\ast ]\times B_{max})/K_{\mathrm{D}}$$ Equation 4

If the ratio of B to free radioligand concentration ([R*]) under these conditions is defined as binding potential (BP) [22], then:

$$BP=B/[\mathrm{R}\ast ]=B_{max}/K_{\mathrm{D}}$$ Equation 5

BP is seen to be the product of the target protein density (B_max) and the affinity (1/K_D) of the radioligand. Therefore, for radioactivity measured in brain with PET to be dominated by radioligand that is not free but specifically bound to target protein, B_max should well exceed the K_D of the radioligand when these are measured in the same units, typically chosen as nM. This is a very important guideline for a successful PET radioligand. Considerations surrounding this guideline are expanded on later (Section 5).

The preceding theoretical discussion pertains to radioligand binding to a target protein at equilibrium in vitro. In PET experiments, the conditions ([R*] ≪ K_D) required for Equation 4 to be applicable are usually met with PET radioligands administered along with only low amounts of carrier i.e., administered at high specific radioactivity, where specific radioactivity is the ratio of ligand radioactivity (Bq) to the total amount of ligand (radioactive plus non-radioactive; mol).

Some PET radioligands are administered intravenously by a bolus injection followed by constant infusion in order to achieve equilibrium conditions over the time-course of scanning. Under these conditions, one version of binding potential is simply the concentration of target-bound radioligand in the brain tissue of interest (C_S) divided by the total concentration of radioligand in blood plasma (free and bound to blood proteins) (C_P). By a quite recent expert consensus [23], this version is termed BP_P, with the subscript P denoting plasma. On the assumption that the K_D of the radioligand for binding to the target protein is constant across brain in vivo, BP_P gives a measure that is directly proportional to the target protein density available in vivo, known as B_avail. In most circumstances, B_avail will be less than the actual B_max value because in vivo protein targets are often partially occupied by their endogenous ligands and/or by other molecular entities, or may be only partially accessible to radioligand.

Other versions of BP pertaining to radioligand binding in vivo are BP_F and BP_ND [21, 23]. BP_F is the ratio at equilibrium of the brain receptor-bound radioligand concentration to that of free radioligand in brain tissue. The latter concentration at equilibrium is assumed to equal that which is free in plasma for a radioligand that only crosses the blood-brain barrier (BBB) by passive diffusion in either direction. Analogously, BP_ND is the ratio at equilibrium of receptor-bound radioligand concentration to that of non-displaceable radioligand concentration in tissue, where the non-displaceable radioligand is the sum of free and nonspecific components. BP_ND may be shown to relate to B_avail, the K_D of the radioligand in vivo, and f_ND, the fraction of free ligand in target-free tissue (the non-displaceable compartment) by the equation:

$$B{P}_{\text{ND}}=f_{\text{ND}}\times (B_{\text{avail}}/K_{\mathrm{D}})$$ Equation 6

The other two versions of binding potential are also directly proportional to B_avail/K_D, and are mathematically defined as follows:

$$B{P}_{\mathrm{F}}=B_{\text{avail}}/K_{\mathrm{D}}$$ Equation 7

$$B{P}_{\mathrm{P}}=f_{\mathrm{P}}\times (B_{\text{avail}}/K_{\mathrm{D}})$$ Equation 8

Equations 6–8 may be appreciated graphically in terms of radioligand distribution in blood and brain at equilibrium (Fig. 2).

Fig. (2).

Graphical representation of the distribution of a PET radioligand between plasma and 2 tissues in brain at equilibrium. The height of each rectangle represents radioligand concentration. The free concentration of radioligand in plasma and brain is assumed to be the same and equal to [A], because of assumed free unhindered bidirectional passive diffusion across the BBB. The plasma free fraction of radioligand, f_P, is the ratio of free radioligand concentration [A] to total radioligand concentration in plasma i.e.; free [A] plus protein bound [B]. Similarly, the brain free fraction of a radioligand, f_ND, is the ratio of free radioligand concentration [A] to the sum of free [A] and nonspecifically bound [C] radioligand concentrations in brain. [D] Represents the specifically bound concentration of radioligand in brain. Some mathematical relationships of binding potentials to other parameters are shown above the right column.

An important assumption underlying measurement of any kind of binding potential is that the radioactivity measured in brain tissue with PET is composed solely of unchanged radioligand; that is without contamination by any radiometabolite. Experimental radioligands may fail to be useful if they generate radiometabolites in brain. The vast majority of PET radioligands are found to metabolize extensively in periphery (see Section 12). Except in some modeling approaches mentioned later, estimation of an output parameter, such as a binding potential, requires determination of the unchanged radioligand concentration in plasma. Analyses are usually performed by separating plasma from arterial blood, removal of proteins by precipitation, and finally chromatographic separation of parent radioligand from circulating radiometabolites. Methods avoiding need for precipitation of proteins from plasma have also been brought into recent practice.

Usually PET radioligands are administered as brief bolus injections, and the time-courses of radioactivity concentrations in brain regions are measured with PET. For many radioligands, the time-course of exposure of the brain to unchanged radioligand in plasma must also be determined, if quantitative output measures are to be obtained from a PET experiment. For this purpose, blood must be sampled at several time-points throughout the scanning period and analyzed for the concentration of unchanged radioligand in plasma to provide the ‘metabolite-corrected arterial input function’. Compartmental modeling [20, 21] may then be applied to the PET and plasma data in order to derive output measures reflecting target protein densities. Mostly, two types of compartmental model are in use, a one-tissue compartmental model (1-TCM), in which brain is considered to be one compartment, and a two-tissue compartmental model (2-TCM) in which brain is considered to have kinetically distinguishable specific and non-displaceable compartments (Fig. 3).

Fig. (3).

One and two tissue compartmental models that are frequently used as a basis for quantification of PET radioligand behavior: (C_P, plasma radioligand concentration; C_T, brain tissue radioligand concentration; C_ND, nonspecific plus free radioligand concentration; C_S, specific radioligand concentration).

In practice, an important output parameter is the total volume of distribution, V_T. Essentially, V_T is defined as the ratio of the concentration of the non-metabolized radioligand in the tissue of interest (C_T) to that in plasma at equilibrium (C_P), i.e.

$$V_{\mathrm{T}}=C_{\mathrm{T}}/C_{\mathrm{P}}$$ Equation 9

For a 1-TCM, it follows that V_T also equals K₁/k₂.

V_T is derived from PET measurements that have a mm-order of spatial resolution. Consequently, each image voxel (volume element) contains radioligand that is: i) free in water space, ii) nonspecifically bound to tissue, and iii) specifically bound to the target protein. (In reality, each voxel also contains radioactivity in blood, but this is sometimes neglected, as blood accounts for only about 5% of human brain volume, or is otherwise accounted for by a fixed or fitted correction). Thus, V_T is the sum of three components, the volume of distribution of radioligand that is: i) free in tissue water (V_F), ii) nonspecifically bound (V_NS), and iii) specifically bound (V_S), i.e.

$$V_{\mathrm{T}}=V_{\mathrm{F}}+V_{\text{NS}}+V_{\mathrm{S}}$$ Equation 10

Because only specifically bound radioligand can be displaced by a competing ligand, the sum of V_F and V_NS is termed the volume of distribution for ‘non-displaceable’ binding, V_ND, i.e.

$$V_{\mathrm{T}}=V_{\text{ND}}+V_{\mathrm{S}}$$ Equation 11

V_ND is usually considered to be constant across brain. Thus, normally, both V_T and V_S reflect the available concentration of unbound target protein, B_avail. In fact, V_S is the same as BP_P. However, V_ND must be determined to obtain V_S. V_T has special importance as a useful output measure for PET studies when V_ND estimates are not easily accessible, as is often the case for human subjects. Clearly, radioligands that give high V_S in one or more target-rich regions relative to V_ND, and hence a high V_T value, are most sensitive to target protein density and will tend to be the most useful in practice. Thus, minimizing V_ND among a series of candidate PET radioligands can be a key component for achieving success in a search for an effective radioligand (see Section 11). It should be noted that with respect to a 2-TCM (Fig. 3), V_ND is simply the ratio of K₁ to k₂, i.e.

$$V_{\text{ND}}=K_1/k_2$$ Equation 12

The binding potential most often measured with PET is BP_ND, which may be expressed in terms of volumes of distribution, as follows:

$$B{P}_{\text{ND}}=(V_{\mathrm{T}}-V_{\text{ND}})/V_{\text{ND}}=(V_{\mathrm{T}}-V_{\text{ND}})-1$$ Equation 13

V_T/V_ND is sometimes called the distribution volume ratio or DVR, and hence:

$$B{P}_{\text{ND}}=\mathit{DVR}-1$$ Equation 14

Measurement of a metabolite-corrected arterial input function requires arterial cannulation of the subject. Avoiding the need for arterial sampling is preferable, if possible, to reduce risk and, more importantly, to reduce resource requirements and simplify methodology. For a few radioligands, the radioligand concentration in blood may be followed with PET at the same time as brain imaging, to provide an ‘image-derived input function’ (IDIF). An example is [¹¹C](R)-rolipram [24], a slowly metabolized radioligand for the phosphodiesterase subtype-4 (PDE4) enzyme. Even so, IDIF still requires a few blood samples to be taken - to ‘calibrate’ the PET measurements on blood.

For certain radioligands, a sizeable region within the brain is known to be devoid of the target protein. Because V_ND is usually quite uniform across brain regions, uptake of radioactivity in this region may be taken to represent non-displaceable binding. This concept underpins useful approaches to quantification of target protein density with PET radioligands without the need for arterial cannulation, such as the ‘simplified reference tissue model’ (SRTM) [25, 26].

Other useful biomathematical approaches have also been developed for deriving quantitative output measures. These include the popular ‘Logan graphical analysis’, which also gives V_T as an output measure [27–29].

Simultaneous PET measurements of both B_avail and K_D are possible. One approach is based on ‘Scatchard analysis’, as often performed in vitro. The radioligand is administered serially with different amounts of carrier to the same subject and the ratio of bound to free radioligand concentration is plotted against bound radioligand concentration. B_avail and K_D are given by the intercept on the X-axis, and the negative magnitude of the inverse of slope, respectively (Fig. S1). Such measurements have been performed with PET in rat [30], NHPs [31], and occasionally in human [32–34]. They are quite logistically demanding because of the necessity for more than one radioligand injection. Hence, only two injections are normally performed for this type of analysis. B_avail and K_D may also be determined from a single radioligand injection [35]. Generally, however, radioligand affinity is assumed to be constant in vivo and the V_T value, obtained from a single radioligand injection, is taken as the index of target density.

The foregoing discussion on the quantification of PET radioligand binding to brain proteins already alludes to the many properties to be sought in candidate radioligands, as highlighted in former reviews. These include a high ratio of target protein B_avail to radioligand K_D in target brain regions, selectivity for the target versus off-target sites, ability to penetrate the BBB following intravenous injection, low nonspecific binding, lack of brain-penetrant radiometabolites, and amenability to labeling with carbon-11 or fluorine-18. Subsequent sections of this article now expand discussion on these and other important considerations in PET radioligand development.

3. REGIONAL TARGET PROTEIN DENSITY

In any program aimed at developing a PET radioligand for a protein target that has not previously been imaged, prior knowledge of the expected concentration of the target is a primary concern. Based on Equation 6, the lower the target density (B_avail) the higher must be the affinity (1/K_D) of a radioligand for successful imaging to be possible.

A priori considerations are whether the imaging target is already expressed in normal human brain, and whether the density is expected to increase or decrease in one or more regions during the progression of a disease of interest. Most proteins that have been targeted for PET imaging have presence and function in normal brain; they include, for example, neurotransmitter transporters, receptors and enzymes. Through the use of PET radioligands, many of these targets have been found to show density changes in the progression of certain neuropsychiatric disorders [2]. Often, these changes are reductions in density rather than increases, and often they are relatively small. The direction of a small change may remain unclear, even after many PET studies as, for example, has been noted for serotonin subtype-1A (5-HT_1A) receptors in depression [36]. The development of higher-performing radioligands may help to resolve such lingering uncertainties, and presents an ongoing challenge to chemists within this field. Gender differences in protein density and also changes with age, usually a progressive decline, are not uncommon [2]. For some other protein targets, substantial increases in density over low basal levels occur with progression of disease. Prominent examples are Aβ plaque in Alzheimer’s disease [37, 38] and the translocator protein 18-kDa (TSPO) in various neuroinflammatory disorders [39–42].

A further consideration is that not all subjects may present the same version of the imaging target. In particular, recent studies of radioligands for imaging TSPO highlight how human genetic variation can be an important factor in PET radioligand performance and utility. Thus, the radioligand [¹¹C]PBR28 succeeds in imaging brain TSPO in only about 90% of human subjects of European ancestry [43]. The failure in 10% of subjects stems from a codominant rs6971 single nucleotide polymorphism (SNP) that causes a non-conservative threonine-to-alanine substitution at position 147 of TSPO protein [44, 45]. This SNP results in three subject groups: homozygous for threonine-147, homozygous for alanine-147, and heterozygous. The relative prevalence of these three groups is about 49: 9: 42 for subjects of Caucasian ancestry. TSPO having alanine in position 147 has very low affinity for PBR28 [46], and this explains why the approximately 10% of individuals that are homozygous for this SNP cannot be imaged with [¹¹C]PBR28 (Fig. 4). Heterozygous individuals do give PET images, but their binding potentials are effectively about half those of threonine-147 homozygous individuals [45]. The three groups are now dubbed high-affinity binders (HABs), low-affinity binders (LABs) and mixed-affinity binders (MABs), respectively. LABs are readily excluded from PET studies with [¹¹C]PBR28 by their inability to give an image of TSPO. However, if MABs are not excluded by genotyping, their presence may reduce the power of PET studies that compare TSPO in diseased and normal brain [45]. Many alternative TSPO radioligands, including those in other structural classes, show some genotype sensitivity and therefore have this issue in common with [¹¹C]PBR28 [47–49].

Fig. (4).

Imaging of TSPO in brain and body organs under baseline conditions in PBR28 binders (HABs) and non-binders (LABs) with the prototypical radioligand [¹¹C](R)-PK11195 and with the later generation radioligand [¹¹C] PBR28. From Kreisl et al. [43], with permission. Unlike [¹¹C](R)-PK11195, [¹¹C]PBR28 shows no TSPO-specific binding in brain or periphery of non-binder.

Considerable evidence exists for genotype sensitivity among radioligands for some other imaging targets, but as yet this evidence is not as comprehensive as for TSPO radioligands. Notably, Willeit and Praschak-Rieder [50] have reviewed the apparent influence of polymorphisms on radioligand binding to sites involved in brain monoaminergic neurotransmitter systems. They found strong evidence for the influence of the triallelic 5-HTTLPPR LA allele on increasing [¹¹C]DASB binding to the serotonin transporter (SERT), and also for the influence of the D2Ta11 A1allele carriers on decreasing the binding of [¹¹C]raclopride to dopamine D₂-like receptors.

Some protein targets are diffusely distributed across human brain, such as the PDE4 enzyme, the cannabinoid subtype-1 (CB₁) receptor, and the metabotropic glutamatergic subtype-5 (mGlu5) receptor. They provide no clear opportunity to use a single region as a target-devoid tissue in PET imaging with a convenient SRTM. Other protein targets, such as dopamine subtype-2 (D₂), serotonin subtype-1A (5-HT_1A) and μ-opiate receptors, and the dopamine transporter (DAT), are more discretely located in one or more sizeable brain regions, and do provide opportunities to apply SRTM. Thus, knowledge of human brain regional target distribution, if only qualitative, is valuable upon entering into a PET radioligand development program.

All too often information on target protein density in normal or diseased human brain is lacking or difficult to obtain. This may be because of a scarcity of suitable post mortem human brain tissue or an absence of a suitable radioligand for in vitro measurements, or both. Ideally, the target protein density across normal human brain regions will have been estimated ahead of PET radioligand development through quantitative brain autoradiography [51–55], or through a binding assay on brain tissue with a selective radioligand [56]. Usually, such measurements are performed with radioligands having a high specific activity tritium or radioiodine label, but they are also sometimes feasible with a radioligand having a carbon-11 [54] or fluorine-18 [55] label.

Protein densities determined in vitro are often cited in units of fmol/g of brain tissue or fmol/mg of protein. For proposed radioligands, B_max values are usefully expressed in units of nM for direct estimation of the in vitro BP (i.e., B_max/K_D) values. Generally, brain is assumed to contain 100 mg of protein per g of wet tissue [57], in which case 1 fmol/g of brain tissue simply equates to 1 nM and 1 fmol/mg protein to 0.1 nM. The range of normal human brain protein densities that has been imaged with PET spans more than two orders of magnitude, for example from about 0.5 nM for the α₄β₂ subtype of nicotinic acetylcholine receptors (α₄β₂nACh) receptors [58, 59] in cortex to over 150 nM for dopamine subtype-1 (D₁) receptors in striatum [60] (Table 2). Estimates of human protein target density are recognized to vary considerably with the type of measurement (e.g., autoradiography in vitro, membrane assay in vitro, or PET) and the choice of radioligand [60]. Therefore, in vitro estimates of binding potentials that are to be expected from PET studies may not be reliable and must be treated cautiously.

Table 2.

B_max, K_D, and B_max/K_D values determined in vitro on normal human brain tissue for some protein targets successfully imaged with PET.

Protein	Brain region	Radioliganda	Bmax (nM)	KD (nM)	Bmax/KDb	Reference
5-HT1A	Parahipp. gyrus	[3H]WAY-100635	24±2	1.1±0.2	22	[61]
5-HT4	Caudate	[3H]GR113808	9±2	0.07±0.01	129	[62]
	Caudate	[3H]GR113808	14.9	0.20	75	[63]
5-HT6	Striatum	[125I]SB-258585	22±4	1.3	17	[64]
α4β2nAChR	Cortex	6-[18F]fluoro-A-85380	0.7	0.059	12	[65]
Aβ plaques	Temporal cortex	[11C]PIB	28b	~2	~ 14	[66]
Bz	Neocortex	[3H]Flumazenil	35	6.1	6	[67]
	Temporal cortex	[3H]Flumazenil	70±2	2.2±0.1	32	[68]
		[3H]Sarmazenil	54±2	1.9±0.2	29	[68]
CB1	Frontal cortex	[3H]MePPEP	47±17	0.3±0.1	157	[69]
D1	Caudate	[3H]SCH23390	103±17	2.1±0.6	49	[70]
D2	Putamen	[3H]Raclopridec	8±1	0.9	9	[71]
	Putamen	[3H]Racloprided	9.9±0.9	3.89	3	[71]
	Caudate	[3H]Raclopridee	24±9	2.5±0.2	10	[72]
DAT	Putamen	[3H]RTI-55	52±17	0.10±0.02	520	[73]
GlyT-1	Cortex	[3H]GSK931145	57±7	4.5±0.8	13	[74]
mGluR5	Cortex	[18F]FPEB	25±6	0.11±0.04	230	[75]
Opiate	Frontal cortex	[3H]Diprenorphine	14.8±0.9	0.22± 0.03	67	[76]
NET	Locus coeruleus	[18F](S,S)-FMeNER-D2	35±3	3.6	10	[55]
SERT	Hippocampus	[3H]Citalopram	28±4	3.6±0.8	8	[77]
NK1	Striatum	[3H]GR205171	43±3	0.02	2150	[78]
TSPO	Temporal cortex	[3H]PK11195	71±6	2.9±0.5	24	[79]

^aLigands shown in bold, when labeled with carbon-11 or fluorine-18, have been used for imaging the target protein with PET.

^bRounded to nearest integer.

^cAD brain

^dat 23 °C

^eat 37 °C

Data on target protein distribution and regional densities are of course more readily accessible from animals than from human subjects. These data must also be used cautiously because major species differences may exist. Such variations should be considered in selecting animal species for the pre-clinical evaluation of prospective radioligands with PET. Ideally, a species with a target density similar to or greater than that in human would be chosen. For example, the density of mGlu5 receptors is much higher in rhesus monkey or human cerebellum than in rat cerebellum [75], and hence monkey is preferred over rat for the evaluation of experimental mGlu5 receptor radioligands. The density of TSPO in normal rhesus monkey brain is at least an order of magnitude higher than in normal human brain, and therefore PET signals obtained with [¹¹C]PBR28 in human are not as strong as those seen in monkey [80]. Nonetheless, rhesus monkey has regularly served as a useful animal model for the pre-clinical evaluation of candidate TSPO radioligands.

The form of the target protein that is possible to measure in vitro may not correspond to what is intended to be imaged in vivo. For example, consider targets that are G-protein-coupled receptors (GPCRs). Measurement of target density in vitro with an antagonist radioligand will measure the density of all available target receptors, regardless of the proportion that is present in the G-protein coupled state at any instant. This information will be most relevant to the development of PET radioligands that are themselves antagonists, but will be less relevant to the development of agonist radioligands intended to bind only to the sub-population of a GPCR target that is actually coupled to G-protein. Brain cryosections that are used for in vitro autoradiography represent a quasi-physiological environment, as do membranes and cells in binding assays [81]. Consequently, these techniques also offer opportunities for identifying proteins in active functional states, such as GPCRs interacting with second messenger systems. Clearly, wherever possible, the measurement of target density in vitro should be relevant to the form of the protein that is intended to be measured in vivo with PET. Other examples of situations in which this consideration would become relevant are in PET imaging of an open channel version of a precursor protein (e.g., the open N-methyl-D-aspartate (NMDA) receptor). However, the expected prevalence of a sub-population of an imaging target relative to the full population is often uncertain because of the absence of in vitro estimates or because of difficulties in making such estimates.

In a few cases, the targeted binding site may only be truly accessible in post mortem brain tissue from human subjects. A pertinent example occurs in the development of radioligands intended to image Aβ plaques in AD. Notably, the use of artificial plaques generated in vitro, and also of plaques generated in transgenic rodents, frustrated early pre-clinical efforts to develop PET radioligands, because of the lack of correspondence in radioligand binding site architecture in these models to that in human brain [82].

A further consideration is that a single target protein may contain more than one type of radioligand binding site. Aβ plaques are again an important example; they present at least three distinct binding sites, which may be classed on the basis of their strong interactions with particular ligand types, namely Congo Red, thioflavin T and FDNNP [82]. Some GPCRs, such as mGlu5 receptors, and also ion channels, such as NMDA receptors, present allosteric as well as orthosteric binding sites, and, in principle, either may be exploited for PET quantification of the target protein.

GPCRs and many other proteins would be expected to have the same number of allosteric and orthosteric binding sites for antagonist radioligands. However, some proteins are known to offer different B_max values for different ligand classes. Aβ plaques are again a notable example [82].

Finally, radioligands that succeed for delineating protein distribution and density in autoradiography or in membrane assays will not necessarily succeed in vivo where confounding effects of poor delivery, metabolism and unacceptable nonspecific binding may readily occur. PET may be thought of as ‘in vivo autoradiography’. Radioligands that fail autoradiography in vitro are virtually certain to fail in vivo.

4. RADIOLIGAND AFFINITY

The foregoing discussion has already raised the importance of radioligand affinity for successful imaging of a protein target. By Equation 6 binding potential (BP_ND) is proportional to affinity (1/K_D). Estimates of ligand affinities for a particular protein are generally derived from an in vitro radioligand competition binding assay employing a reference radioligand, R*, that binds tightly but reversibly to the target protein. Essentially, the test ligand is incubated at different concentrations in the assay medium along with the radioligand and target protein. The binding of the radioligand to the protein is determined at equilibrium at each test ligand concentration, resulting in a competition binding curve (Fig. S2). Assays are designed to provide the protein in a particular biological matrix, which may be tissue homogenate, isolated membranes, or intact cells [81, 83]. Cells may be produced to over-express the target protein of interest to provide an adequate concentration for use in the assay. In many cases, the human target protein may be usefully expressed in such a manner. The radioligand is generally labeled in high specific activity with a β-emitting radionuclide, often tritium (t_1/2 = 12.32 y), but sometimes radioiodine (e.g., ¹²⁵I; t_1/2 = 59.4 d). Occasionally, a radioligand labeled with a short-lived positron-emitter (¹¹C or ¹⁸F) has been deployed [84, 85]. The output of these assays is an IC₅₀ value, which is the concentration inhibiting 50% of the specific binding of the reference radioligand. Assuming that the K_D value of the reference radioligand is known, for example, from a self-competition assay, the inhibition constant, K_I, for a test ligand can be estimated according to Equation 15, where [R*] is the total concentration of reference radioligand in the assay:

$$K_{\mathrm{I}}=I{C}_{50}/(1+([\mathrm{R}\ast ]/K_{\mathrm{D}}))$$ Equation 15

Consequently, if [R*] is much less than K_D, then K_I approximates to the IC₅₀ value. K_I and IC₅₀ values can be used as surrogates for K_D to rank the binding affinities in a series of test ligands for the target protein, with their inverses representing binding affinity (as earlier explained for 1/K_D). Equation 15 is a version of the well-known Cheng-Prusoff equation [86], originally developed with regard to enzyme-inhibitor interactions. For inhibitors binding to enzymes, the concentration causing 50% inhibition is analogously termed the EC₅₀ value.

Binding affinities determined from competition binding assays may vary with various parameters, including not only the protein matrix, but also temperature and buffer composition [87]. Assays are typically run at 4 or 20 °C. As a result, binding affinities may be unpredictably different from those occurring at mammalian physiological temperature (37–39 °C) in PET experiments. For example, the benzodiazepine receptor radioligand, [¹¹C]flumazenil, has five-fold lower affinity at 37 °C than at 4 °C. This finding partially explained an earlier apparent discrepancy between in vitro and in vivo measures of binding affinity [67]. Similarly, the radioligand [¹⁸F]fallypride was found to have about five-fold lower affinity for rat D₂ receptors at 37 °C than at 22 °C [88]. By contrast, only one of four tested PET radioligands for SERT, showed a significant decrease in affinity between 22 and 37 °C [14].

Attention to buffer compositions is particularly important when testing agonists, if binding affinity values relevant to conditions in vivo are to be obtained. Given these considerations, binding affinities determined at different laboratories often vary considerably, sometimes by an order of magnitude or so. Replications of determinations in the same laboratory by the same methodology may also show some variation due to intrinsic errors in determining inhibition curves. These curves are conveniently plotted as in Fig. (S2), as % inhibition versus log of the test ligand concentration, with concentration typically varied over several decades to obtain the full binding curve. The IC₅₀ value representing 50% inhibition will be at the inflexion in this log-linear curve and may not always turn out to have the desired precision. Possible uncertainty in binding affinity is an important consideration to take into account when selecting candidate PET radioligands for evaluation given that binding potential is linearly proportional to binding affinity, and not to its logarithm. Nonetheless, because binding affinities may occur over a wide range of values, they are often conveniently recorded as the negative logarithm to base 10 of the measured affinity parameter, e.g., as pK_D, pK_I, or pIC₅₀.

GPCRs are often targets for new therapeutics and for PET radioligand studies. GTPγS binding assays have gained popularity for assessing ligand potency and efficacy at GPCRs [89]. These assays are based on assessing the influence of the test ligand on the binding of either [³⁵S]GTPγS or Eu-GTPγS to GPCRs in tissue. For antagonists, GTPγS binding assays can provide K_D values. For agonists and inverse agonists, data on the concentrations giving maximal effect and half maximal response may be obtained. It should be noted that such functional assays do not give direct estimates of agonist binding affinity. This is because the receptor response depends on both the occupancy of the receptor by the agonist and the efficacy of the generated receptor-agonist complex. Also, in some cases, full occupancy of the receptor by an agonist may not be necessary to elicit full receptor response, whereas some agonists may show lower potencies than expected from their affinities as determined in competition binding assays. Other functional assays are also available for GPCRs [90], such as the Ca²⁺ flux assay, which is especially useful for high throughput ligand screening, but again much less so for predicting radioligand affinity. Technology for performing receptor-ligand binding assays also continues to advance, driven by needs for high throughput screening, and includes several non-radiometric methods [91].

It is not unusual for ligand binding affinities to vary considerably across species, particularly for those between rodents and NHPs. Binding assays for a wide range of protein targets are now offered to investigators by the NIMH-sponsored Psychoactive Drug Screening Program (PDSP; https://pdspdb.unc.edu.pdsweb), and also by commercial entities. Assays offered by PDSP are mainly based on recombinant human receptors.

The affinities of PET radioligands for their targets have kinetic consequences. In general, the higher the affinity of the radioligand (with all other factors being equal), the slower will be the brain kinetics following a bolus injection, as represented by a slower peak radioactivity in a target-rich brain region and a slower washout. This is due to the longer time needed to reach a transient equilibrium between radioligand association and dissociation, as has been shown by simulation [14]. Over-slow kinetics is undesirable because of the consequent need for a long PET scan duration to acquire sufficient data to generate robust quantitative outputs. Studies on radioligands for D₂-like receptors illustrate this point. [¹¹C]Raclopride (K_D = 3.89 nM) has long been used to measure D₂-like receptors in human striatum. The structurally related benzamide [¹¹C]FLB 457 has much higher affinity (K_D = 20 pM) for D₂-like receptors, and is suitable for measuring the much lower densities of these receptors in extrastriatal regions [92]. Transient equilibrium occurs within ten minutes of scanning after bolus radioligand administration. However, kinetics in receptor-rich striatum are very slow such that transient equilibrium is not reached, even after more than one hour of scanning, thus precluding their robust measurement. Slow kinetics has been notably problematic in the development of high-affinity radioligands for sparse α₄β₂nACh receptors [93].

5. BINDING POTENTIAL

Earlier, a simple theoretical basis was given for requiring in vitro estimates of binding potential (BP = B_max/K_D) to well exceed unity in order for a PET radioligand to be successful. In practice, BP values predicted from in vitro measurements of B_max and K_D that are greater than 5, or even much higher, are generally found to be needed (Table 2). This may be appreciated by comparing the data presented in Tables 2 and 3 for [³H/¹¹C]flumazenil, [³H/¹¹C]SCH23390, [³H/¹¹C]raclopride, and [³H/¹¹C]diprenorphine for benzodiazepine (Bz), D₁, D₂-like, and opiate receptor targets, respectively. The complexity of the environment in vivo relative to the highly controlled and simpler environment used for measurements in vitro account for such differences. In particular, the tissue free fraction in brain (f_ND) is a major factor in determining BP_ND, the ratio of specific to non-displaceable binding at equilibrium in vivo. In fact, BP_ND is proportional to f_ND as earlier defined by Equation 6, and illustrated in Fig. (2).

Table 3.

Examples of PET studies determining brain protein density (B_avail) and/or radioligand (K_D) affinity in normal human subjects, and the corresponding B_avail/K_D values.

Protein	Brain region	Radioligand	Bavail from PET (nM)	KD from PET (nM)	Bavail/KD from PETa	Reference
Bz	R. m. t. ctx Occ. ctx.	[11C]Flumazenil	55±12	12±2	5	[34]
		[11C]Flumazenil	89	13.7	7	[94]
D1	R. caudate	[11C]SCH23390	45±12	22±8	2	[95]
D2	Putamen	[11C]Raclopride	14±2	3.8±0.6	4	[32]
	R. putamen Striatum	[11C]Raclopride	28± 8	9±2	3	[96]
		[11C]Raclopride	29±4	9±1	3	[97]
D2	Thalamus	[11C]FLB-457	2.3±0.6	0.7±0.3	3	[98]
Opiate	Thalamus	[11C]Diprenorphine	19±2	0.7±0.2	25	[33]

^aRounded to nearest integer.

Ordinarily, f_ND is not directly measureable. However, the non-displaceable volume of distribution V_ND is frequently measureable, especially in animal PET experiments where specific binding of the radioligand has been prevented by a competing non-radioactive ligand (blocking agent) given at a target-saturating dose. With respect to a 2-TCM, V_ND, earlier defined as K₁/k₂ (Equation 16), may also be expressed simply as the ratio of f_P to the brain free fraction (f_ND), for a radioligand that crosses the BBB by passive diffusion only, i.e.

$$V_{\text{ND}}=f_{\mathrm{P}}/f_{\text{ND}}$$ Equation 16

Consequently, BP_ND may also be defined according to the equation:

$$B{P}_{\text{ND}}=(f_{\mathrm{p}}\times B_{\text{avail}})/(V_{\text{ND}}\times K_{\mathrm{D}})$$ Equation 17

This equation theoretically implicates three radioligand properties in determining the magnitude of BP_ND, namely affinity in vivo (1/K_D), non-displaceable binding (V_ND), and plasma free fraction (f_P). As will be discussed later (Section 10), the latter parameter tends to increase with radioligand lipophilicity, as does f_ND. However, for individual radioligands, f_P and f_ND may differ widely because of the different chemical make-ups of blood and brain tissue environments. Data from Guo et al. [99] for the f_P and f_ND values for 28 radioligands in pig clearly illustrate this lack of correspondence (Fig. 5).

Fig. (5).

Plot of f_ND versus f_P for 28 ligands in pig. Data are sourced from from Guo et al. [99]. Although the plot shows a quite strong linear correlation, slope is far from unity, and differences between parameters can be large for many ligands.

The influence of f_p and f_ND on PET BP_ND has been well exemplified by considering the binding of [¹¹C]PIB to Aβ plaques in AD [16]. BP_ND for this radioligand from PET studies is about 2, whereas in vitro the radioligand K_D is about 2 nM and the B_max is estimated to be 1–2 μmol, giving a B_max/K_D value of 500–1000. f_p is known to be < 0.01. By Equation 6, it may be deduced that f_ND is also extremely low (< 0.005), thereby accounting for the several hundred-fold mismatch between B_max/K_D and measured binding potential. Several candidate PET radioligands have likely failed because of the marked effect of low f_ND on BP_ND, even where the in vitro B_max/K_D value appears substantially above unity [16].

In practice, in addition to low f_ND, other factors may also act to decrease apparent BP_ND in vivo, for example, undesirable nonspecific binding from radiometabolites [100], and for brain regions that are quite small relative to the spatial resolution of the PET camera leading to a ‘smearing’ of the radioactive signal through a ‘partial volume effect’.

6. SELECTIVITY FOR TARGET

Clearly, an ideal PET radioligand would bind only to the target protein in brain. Nevertheless, in practice some radioligands show an appreciable affinity for one or more off-target sites, in addition to an inevitable component of nonspecific binding. This raises the question: what level of selectivity for the target site is required in a PET radioligand? There are two main considerations in answering this question. The first is the anatomical location of the competing ‘off-target site’. If this is distinct and distant from the target site, a lack of selectivity may be tolerable. Thus, the D₁ receptor radioligand, [¹¹C]NNC 112, also has high affinity for 5-HT_2A receptors. This is problematic for imaging D₁ receptor in cerebral cortex where about 25% of specific binding is to prevalent 5-HT_2A receptors. This radioligand is nonetheless useful to measure D₁ receptor in striatum, because of their high density relative to the low density of 5-HT_2A receptors [101].

If the off-target site and the target site are found in the same region of interest, binding of the radioligand to the off-target site relative to that of the target site must be considered. The BP_ND measured with PET will be the sum of the binding potentials at each site. On the basis of Equation 6, this sum may be expressed as:

$$B{P}_{ND}={{}^{\mathrm{O}}f}_{\text{ND}}({{}^{\mathrm{O}}B}_{\text{avail}}/{{}^{\mathrm{O}}K}_{\mathrm{D}})+{{}^{\mathrm{T}}f}_{\text{ND}}({{}^{\mathrm{T}}B}_{\text{avail}}/{{}^{\mathrm{T}}K}_{\mathrm{D}})$$ Equation 18

in which the prefixed superscripts ^O and ^T denote parameters for the off-target site and target site, respectively. By the likely reasonable assumption that the same free fraction of radioligand (f_ND) is available to each type of site, Equation 18 simplifies to:

$$B{P}_{ND}=f_{\text{ND}}({{}^{\mathrm{O}}B}_{\text{avail}}/{{}^{\mathrm{O}}K}_{\mathrm{D}}+{{}^{\mathrm{T}}B}_{\text{avail}}/{{}^{\mathrm{T}}K}_{\mathrm{D}})$$ Equation 19

Therefore, the relative B_avail/K_D values of the radioligand for the target and off-target site need to be considered, usually on the basis of in vitro estimates of B_max/K_D.

An example of two target proteins that have been of interest for PET imaging and that show overlapping anatomical distribution are DAT and SERT. Both DAT and SERT are expressed in the striatum, but DAT is by far the more abundant. Consequently, DAT can be imaged with PET by using radioligands that have quite low affinity and quite low selectivity for binding to DAT versus SERT. However, by contrast, the imaging of SERT requires a radioligand with both high binding affinity and very high SERT selectivity [102].

Recent efforts to develop radioligands for neurofibrillary tangles (NFTs) in tauopathies provide another interesting example. Here it is important that the radioligand is selective for binding to NFTs versus Aβ plaques, which may also be present in these and other neurodegenerative disorders at greater abundance than NFTs. By considering density estimates of these proteins, Schafer et al. [103] estimated that a useful NFT-selective radioligand would need to show more than 20-fold greater affinity for NFTs than for Aβ plaques. Radioligands that show selectivity for imaging NFTs over Aβ-amyloid plaque are now beginning to emerge.

Given that the monoamine transporters, DAT, SERT and NET share a high degree of structural homology, it is not surprising that the development of selective high-affinity radioligands has been a challenging task. Likewise, both NFTs and Aβ plaques display extensive β-sheet structures that are important for ligand binding, and hence this similarity is a challenge in developing selective radioligands. More surprising is the lack of selectivity of some high-affinity ligands for binding to structurally dissimilar proteins. For example, some high-affinity ligands for the CB₁ receptor, a seven transmembrane domain GPCR, were also found to bind strongly to TSPO, a channel composed of five subunits [104]. Despite the strong structural dissimilarity in these proteins, their pharmacophore requirements appear to be quite similar with regard to the gross distribution of H-bonding and lipophilic groups.

The initial assessment of radioligand selectivity is usually made on the basis of in vitro screening for affinities against a broad battery of receptors and binding sites, as offered by PDSP and commercial entities. The selectivity of a radioligand may also be checked with PET in a suitable living animal model, preferably NHP. Challenge of the PET signal by simultaneous or preceding administration of the non-radioactive ligand in high dose is sometimes very useful to show whether appreciable specific binding of any kind is occurring in brain. Such experiments should be done rigorously to account for possible differences in brain exposure to radioligand in plasma, for example, by application of compartmental modeling with a metabolite-corrected arterial input function to determine any changes in V_T, or preferably V_T/f_P, under baseline and challenge conditions. Such an experiment leaves open the question of whether any observed specific binding is associated only with the target binding site. To answer this question, a compound which is known to be selective for the target site should be given at high dose before the radioligand. This compound should ideally belong to a different structural class, be safe for administration at high dose, and be expected to occupy all target sites at the time of radioligand injection. In some cases this type of administration may increase the peak uptake of radioactivity seen in brain, because blockade of peripheral target of the same type as the brain target may increase radioligand availability in plasma. This is commonly the case for experiments performed with TSPO-selective radioligands [48, 105, 106], and also for SERT radioligands which, may have avid binding in lung. When it is important to check that the radioligand is not binding to a particular off-target site, the PET signal may be challenged with a ligand that is selective for that off-target site. An alternative to administering challenge ligands preceding radioligand administration is to perform a displacement experiment, in which the challenge agent is given after the radioligand, and to observe whether radioactivity is then more rapidly cleared from target regions or not. For example, the selectivity of the radioligand [¹⁸F]FIMX for binding to mGlu1 versus mGlu5 receptors has been checked in this manner (Fig. 6) [107].

Fig. (6).

PET Time-activity curves in monkey brain regions after administration of [¹⁸F]FIMX. (A) At baseline; (B) after pre-block with a selective mGlu1 receptor antagonist; (C) at baseline with challenge from an mGlu1 receptor antagonist at 27.5 min; and (D) at baseline with challenge from an mGlu5 receptor antagonist at 27.5 min. (Adapted from [107]). The experiments show that [¹⁸F]FIMX has high specific binding to mGlu1 receptors at baseline.

7. LIGAND EFFICACY (INTRINSIC ACTIVITY)

The concept of ligand efficacy (intrinsic activity) is at the core of pharmacology. As a recap there are various classes of ligand efficacy. Ligands that bind to receptors may elicit a full pharmacological effect and are termed full agonists. Those that fully occupy receptors without eliciting a full pharmacological effect are termed partial agonists. Ligands that diminish the agonist-independent activity of a receptor, or that reverse a pharmacological effect either fully or partially to another type, are full or partial inverse agonists, respectively. Ligands that bind to the receptors without eliciting any function and which are capable of blocking the effect of all types of agonists are termed antagonists. These various terms may be exemplified by the pharmacological actions of various ligands for brain GABA_A (Bz) receptors [108] (Fig. 7). Antagonists, such as flumazenil (Ro-15-1788), are without pharmacological effects, whereas full agonists, such as diazepam, cause sedation, and full inverse agonists, such as Ro-19-4603, cause seizures. Partial agonists, such as bretazenil (Ro-16-6028), are anticonvulsant, and partial inverse agonists, such as sarmazenil (Ro-15-3505), are anxiogenic [109]. It is important to note that ligands may not display the same type of efficacy in all situations in vitro. Moreover, many ligands that have originally presented as antagonists have turned out to be inverse agonists on deeper investigation [110]. Nor does ligand efficacy relate to ligand affinity; for example some partial agonists may have higher binding affinities than full agonists.

Fig. (7).

Illustration of the spectrum of ligand efficacy with ligands for benzodiazepine receptors.

A high proportion of PET radioligand development for brain imaging has been directed at GPCRs and other functional neurotransmitter receptors. Leads for such radioligand development have frequently emerged from medicinal chemistry efforts to develop either agonists or antagonists as drugs. Before proceeding to discuss how radioligand efficacy at GPCRs affects PET radioligand imaging performance it is useful to give some background with regard to how GPCRs are now thought to function. Some of the earliest models with utility for understanding GPCR function are so-called ‘general two-state models’, in which a GPCR is considered to be in equilibrium between non-activated (Re) and activated states (Re*) [111] (Fig. 8). In such models, an agonist (Ag) binds with these forms of the receptor to alter the position of equilibrium towards Re* and thereby generate a cellular response. Inverse agonists are considered to shift the equilibrium away from Re*. In a further refinement of this model, known as the ‘extended ternary complex model’, it is considered that both Re* and the complex with agonist AgRe* must become bound reversibly to G-protein to generate a cellular response (Fig. 8). According to such models, agonists bind with high affinity to Re*G, the G-protein coupled form of Re*, and with lower affinity to the non-G-coupled form Re. In relatively early studies, ratios of affinity for the G-protein-coupled state (1/K_Dhigh) to affinity for the uncoupled state (1/K_Dlow) in vitro were found to correlate well and linearly with ligand efficacy (E%) at several GPCRs, such as 5-HT_1A [112,113], 5-HT_2A [114], 5-HT_2C [114], CB₁ [115], and D₂ receptors [116]. A low E% value would correspond to a partial agonist and 100% efficacy to a full agonist. In these studies, full agonists showed affinity ratios that depended on the particular GPCR, and they ranged from 11.7 for serotonin at 5-HT_2A receptors to 238 for dopamine at D₂ receptors.

Fig. (8).

The general two-state and extended ternary complex models of GPCR function induced by an agonist. Ag is agonist, Re is receptor, Re* is receptor in activated state, and G is a G-protein.

Evidence that has been accumulated from studies of GPCR functional selectivity, biophysics and crystal structures now shows that the early models of GPCR function are too simple. GPCRs are now known to adopt a variety of interconverting conformational states, whether free or bound to ligand or to a G-protein or both [117–119]. Each conformational state is associated with a particular free energy. This gives rise to the notion of an energy landscape relating all free and bound forms of the receptor. Conformational intermediates that are involved in receptor activation or deactivation can be regarded as wells in the energy landscape, and the deeper any particular well the more stable is the conformation. Ligand structure determines the particular make-up of the conformational ensemble that will be bound, and therefore these ensembles vary from ligand to ligand. Ligand efficacy for different signaling pathways derives from the ligand stabilization of particular conformational ensembles that interact with specific effectors. Thus, agonists are considered to bind with high affinity to only a sub-set of relatively high-energy receptor conformations. Such binding lowers the energies of the bound conformations and increases their relative concentrations. The energy landscape concept has become especially helpful in understanding how GPCRs may recruit different G-proteins for activation of different pathways that lead ultimately to different cell responses. This concept also helps to explain the basal or constitutive activity of most GPCRs. Thus, GPCRs are no longer viewed as simple ‘two-state switches’, but more like ‘molecular rheostats’ that present an array of conformations with relatively close energy differences.

Overall, antagonists have been the most productive leads for developing effective PET radioligands for GPCRs. This is at least partially explicable on the basis that the density of a protein target (B_avail) that is available to an antagonist will be at least as high as that for an agonist because antagonists bind with similar affinity to all versions of the target receptor in both G-protein coupled and non-coupled states. The history of PET radioligand development for brain 5-HT_1A receptors illustrates how ligand efficacy may influence the ability of radioligands to image GPCRs. Many early efforts preceded the development of high-affinity antagonist radioligands for 5-HT_1A receptors and therefore were obliged to be based on agonists or partial agonists. Despite the availability of agonist leads with very high affinity [120], effective PET radioligands giving high specific to nonspecific binding signals in rodents or NHPs failed to emerge. Only upon the development of antagonist radioligands based on the ‘silent’ high-affinity antagonist WAY-100635 were 5-HT_1A receptors first imaged and quantified effectively in human brain [120–124].

For further illustration, Fig. (9) plots pA (negative log₁₀ of affinity, where A is IC₅₀, K_I or K_D) versus lipophilicity, as represented by clogD^7.4, for a set of radioligands that have been evaluated for imaging brain 5-HT_1A receptors with PET in rodents or monkeys. Antagonists are plotted as green symbols and non-antagonists as red symbols. Antagonists in the top left-hand sector of Fig. (9) bound by pA > 8.5 and clogD^7.4 < 3.5 give sizeable receptor-specific PET signals. Many of these have progressed to regular use for PET imaging in human subjects, including [¹¹C]WAY-100635, [¹¹C]DWAY, [¹⁸F]FCWAY, and [¹⁸F]MPPF. By contrast, non-antagonist radioligands with very similar parameters fail to give sizeable PET signals. The enantiomer pair, [¹¹C](R)-RWAY and [¹¹C](S)-RWAY, provide an especially interesting comparison. They have similar affinities, identical lipophilicities and virtually identical nonspecific binding levels in receptor-poor cerebellum. The R-enantiomer, an antagonist, gives a sizeable receptor-specific signal in monkey, whereas the somewhat higher affinity S-enantiomer, a partial agonist, gives a much lower signal [127]. Absent or low PET signals from non-antagonists compared to those from antagonists with comparable affinity and lipophilicity supports the notion that only relatively small sub-populations of 5-HT_1A receptors are available for binding to non-antagonists. In this regard, it is interesting to note that the proportion of 5-HT_1A receptors that exists in either a high-affinity G-protein coupled state or a low-affinity G-protein uncoupled state in rat hippocampal membranes is about the same [128].

Fig. (9).

Plot of affinity (pA) versus computed lipophilicity (clogD) for antagonist and non-antagonist radioligands for brain 5-HT_1A receptors. All 26 radioligands have been tested in monkey (solid symbols) or in rat (open symbols). Antagonists shown by green solid symbols and within the top left-hand sector bounded by the line gave strong performance in monkey, whereas agonist radioligands (solid red symbols) with similar affinity and lipophilicity give small or negligible signals. Similarly agonists perform poorly in rat (open red circles). Radioligands 1 (originally classed as an agonist), 4, 7, 8, 9, 14 and 20 have been used in human studies. The plot indicates that high affinity, moderate lipophilicity and antagonist pharmacology are desirable properties for obtaining sensitive radioligands for imaging brain 5-HT_1A receptors in animal or human brain. clogD was calculated with Pallas software. Other data are from reference [120], and references therein, except for radioligands 1 [125], 4 [126], 11 [127], and 12 [127].

A further consideration is that the binding of an agonist radioligand to a GPCR target in a G-protein coupled state in vivo might well be transient, as agonist binding is generally expected to decouple the G-protein rapidly and to generate a target protein conformation with lower affinity [117–119]. Nonetheless, considerable efforts continue to be expended on attempts to develop full agonist-type PET radioligands for two main reasons. Firstly, such radioligands might image only sub-populations of the target receptor that are in functional G-protein-coupled states. Secondly, agonist radioligand binding is expected to be more sensitive to the synaptic concentration of endogenous agonist. Thus, true full agonist-type radioligands might be more selective than antagonist-type radioligands for investigating pharmacological and other influences on neurotransmission.

Continuing efforts to develop agonist radioligands for imaging the high-affinity G-protein-coupled states of the 5-HT_1A receptor have so far met with limited success. For a time, the most promising of these radioligands appeared to be [¹¹C]CUMI-101 [129], but on deeper investigation this radioligand is now considered to be an antagonist in native rat brain tissue [130] and in primate brain [131]. [¹¹C]Raclopride has long served as an effective antagonist radioligand for imaging D₂-like receptors. Extensive efforts have also been directed at the discovery of agonist-type PET radioligands for D₂-like receptors in order to explore dopaminergic neurotransmission. [¹¹C]MNPA has become the preferred and most prominently explored agonist radioligand for this purpose [132]. This radioligand appears to be more sensitive than [¹¹C]raclopride to displacement by endogenous dopamine [133]. However, it is not yet conclusively demonstrated that [¹¹C]MNPA specific binding in vivo, or that of any other agonist PET radioligand, is to high-affinity states of the D₂-receptor [132–135]. In one PET study, an agonist-type radioligand [¹¹C]NPA and an antagonist radioligand, [¹¹C]raclopride, showed similar B_avail values for D₂-like receptors in baboons [136]. A similar result was found in cat when comparing the agonist radioligand [¹¹C](+)PHNO with [¹¹C]raclopride [137]. In this regard, Seeman [138] has argued that [¹¹C](+)PHNO, likely reports ‘low-affinity’ G-protein uncoupled states in vivo, because agonist binding to the receptor results in very rapid sub-second dissociation of the G-protein followed by a not so fast reversion of the receptor to the low affinity state while the agonist remains bound. Therefore, considerable uncertainty surrounds the identity of conformational ensembles that may be imaged with non-antagonist PET radioligands for GPCRs. Different PET radioligands may be imaging different conformational ensembles, and these ensembles may well be mainly composed of non G-protein coupled conformations.

Notwithstanding considerations with regard to agonist radioligands for 5-HT_1A and D₂-like receptors, some agonist radioligands for other GPCRs have been found effective, including, for example, [¹¹C]carfentanil for imaging μ-opioid receptors [139]. The inverse agonist [¹⁸F]FMPEP-d₂ has also been used successfully for imaging abundant brain cannabinoid subtype-1 (CB₁) receptors [140]. The conformational ensembles being imaged by such agonist or partial agonist radioligands again however remains open to question. Some non-antagonist radioligands are also used successfully for PET imaging of non-GPCRs, such as the partial inverse agonist [¹¹C]Ro-15-4513 for GABA_A receptors [141].

The majority of PET radioligands for imaging proteins interact at sites that bind the endogenous ligand and these are termed orthosteric sites. Radioligands may also be developed to interact with sites that are remote from the orthosteric site, termed allosteric sites. Ligands binding to allosteric sites are also classified according to their effects on the action of endogenous ligand. Positive allosteric modulators (PAMs) potentiate response to the endogenous agonist, negative allosteric modulators (NAMs) inhibit response, and silent (or neutral) allosteric modulators (SAMs) are without effect and also block the actions of PAMs and NAMs. The earlier mentioned ligands for GABA_A receptors (Fig. 7) bind to allosteric sites. Allosteric binding sites are often less structurally conserved than orthosteric sites among closely related receptor subtypes, and have become attractive targets for the development of drugs with high binding selectivity [142, 143]. Similarly, they have become of increased interest for developing target subtype-selective PET radioligands. The recent successful development of selective radioligands for PET imaging of mGlu5 receptors [144] and also of very structurally related mGlu1 receptors [107], each based on subtype-selective allosteric modulators, exemplifies the value of this approach.

Finally, a major consideration in the use of an agonist radioligand is its pharmacological action. The mass doses of carrier that are co-administered with such radioligands often need to be limited for subject safety. Antagonist radioligands generally avoid pharmacological concerns.

8. ABILITY TO PASS THE BLOOD-BRAIN BARRIER

Self-evidently, PET radioligands administered intravenously for imaging proteins within the brain must be able to penetrate the BBB and have a high peak uptake that allows for accurate radioactivity measurement. A useful parameter for quantifying radioactivity concentrations in regions of brain or other organs after administration of a PET radioligand is the standardized uptake value (SUV), which normalizes measured concentrations (Bq/g) for injected dose (Bq) and subject body weight (g). SUV may be defined as follows:

$$\text{SUV}=[\%\text{injected}\text{dose}\text{per}\mathrm{g}\times \text{body}\text{weight}(\mathrm{g})]/100$$ Equation 20

An SUV value of 1 in a particular organ or tissue corresponds to the radioactivity concentration that would be expected if administered radioactivity were evenly distributed throughout the subject. Higher SUV values therefore indicate an enrichment of radioactivity concentration above the overall average. The SUV parameter usefully enables comparisons of the radioactivity uptake between different tissues, organs and species. For example, assuming that a typical laboratory mouse weighs 40 g, a radioactivity concentration of 5% injected dose per g of brain would be 2 SUV. For a laboratory rat weighing 400 g, a radioactivity concentration of only 0.5% injected dose per g of brain gives the same SUV value. Successful PET radioligands mostly exhibit peak brain radioactivity values in the range 2–10 SUV (Table 4). A major hurdle to achieving such concentrations of radioligands in brain can be the BBB.

Table 4.

20 PET radioligands: peak brain uptakes in monkey and human, and some of their molecular properties.

Radioliganda	Structure	Target	Peak monkey brain uptakeb [SUV]	Peak human brain uptakeb [SUV]	MWtc [Da]	LogD7.4d	cLogP	N + O	HBDse	HBAsf	PSAg [Å2]	Most basic siteh pKa
[11C](R)-PK11195		TSPO	1.6 [43]	1.6	353	3.97	4.89	3	0	3	33	3.53
[11C]ER176		TSPO	2.0 [145]	2.3 [146]	353	3.55 [145]	3.80	4	0	1	45	0.2
[11C]SD5024		CB1	2.0 [147]	2.3 [147]	463	3.79 [147]	3.50	7	1	2	112	4.98
[11C]IGA-1		TSPO	2.3 [48]	n.d.	406		3.35	7	0	3	92	NA
[11C]PBR28		TSPO	2.8 [105, 148]	2.6 [80, 42, 149, 150]	348	3.01	2.56	5	0	5	51	6.58
[11C]PBR01		TSPO	3.0 [105, 151]	n.d.	374	3.90	3.05	4	0		56	4.4
[11C]RX-1		5-HT1A	3.5 [152]	n.d.	321	1.77	2.36	5	1	3	65	8.47
[18F]FBR		TSPO	3.7 [106]	1.3 [150, 153]	395	4.05	3.58	5	0	5	48	4.99
[11C]RWAY		5-HT1A	4.2 [127, 154, 155]	2.2 [156]	436	3.16	2.85	5	0	5	36	7.94
[11C](R)-rolipram		PDE4	4.3	2.3 [157, 158]	275	2.16	2.35	4	1	3	48	7.61
[18F]RX-2		5-HT4	4.31 [152]	n.d.	289	1.54 [152]	1.42	4	1	3	56	8.47
[18F]XB-1		H3	4.5 [159]	n.d.	351	2.95	4.34	3	0	3	30	9.29
[11C]NOP-1A		NOP	4.9 [160, 161]	6.2 [162, 163]	408	3.41	3.47	4	1	3	42	7.36
[18F]FIMX		mGlu1	5.3 [107]	4.8	343	2.52 [107]	0.69	7	1	3	69	4.95
[18F]FMPEP-d2		CB1	5.5 [140]	4.8 [140]	473	4.24	4.99	4	1	3	42	6.86
[11C]MePPEP		CB1	6.0 [164]	4.0 [164]	455	4.80	4.87	4	1	3	42	6.86
[11C]CUMI-101		5-HT1A	~ 6 [155]	N.A.	373	1.13	0.55	8	0	8	69	9.13
[18F]SP203		mGlu5	7.0 [165]	5.9 [166]	260	2.18	3.66	2	0	1	37	NA
[18F]FECNT		DAT	7.5 [167]	7.5 [167]	325	2.64	3.47	3	0	3	30	6.99
[18F]Fallypride		D2	~ 8.0	~ 10	364		3.33	5	1	4	51	8.70
Criteria:
[168]							> 1.5 & < 2.7
[169]					< 450						< 90
[170]					≤ 400		< 5		≤ 3	≤ 7
[171]											< 70
[172]								≤ 5
[172]								> logP
[173]					< 500	2–5	2–5		< 3		< 90
[174]					< 400		< 4.0
(MDR + LDR) (I) [175]					< 351	< 2.8	< 4.0		< 2		32–86	< 9.5

^aListed in ascending order of peak brain uptake in monkey. Unreferenced values are from the author’s laboratory.

^bWeighted mean of reported values, or of reported range for reported region of highest uptake. Unreferenced values are from the author’s laboratory.

^cFor non-labeled compound.

^dFrom reference [176], unless otherwise indicated.

^eCounted as hetero atoms having a lone electron pair.

^fCounted as electronegative atoms bonded to H.

^gFrom ChemDraw.

^hFrom Pallas software.

The BBB is a complex structure that operates at the capillary level to separate circulating blood from brain extracellular fluid in the CNS. The BBB protects the brain from infectious agents (e.g., bacteria, viruses) and possibly harmful exogenous substances, which may also include small drug-like molecules and radiotracers [177]. The capillary walls of the BBB are formed from endothelial cells that form very tight junctions (Fig. 10). These cells are mainly composed of phospholipid bilayers, having hydrophilic phosphate heads and lipophilic tails (Fig. 11). Consequently, the membrane is generally impervious to passive diffusion by charged or highly polar species, but not to relatively lipophilic species. Free diffusion across the BBB is mainly limited to small molecules. Some small polar species may cross the BBB by an intercellular pathway at the tight junctions, but this pathway is usually negligible. In general, charged molecules are excluded from passage across the BBB, except those needed by the brain for which there are specific transporter proteins, such as those known for essential amino acids and nucleosides. In addition to having these transporter proteins for delivery of specific compounds to the brain, the BBB contains efflux transporter proteins, which actively exclude many unwanted compounds. Many drug-like molecules that might be expected to cross the BBB by passive diffusion, including experimental therapeutics and candidate PET radioligands, are in fact prevented from doing so by the action of efflux transporters, as is discussed later (Section 9). For radioligands that are not substrates for efflux transporters, peak uptake in brain is influenced by many factors, some that stem from the physicochemical properties of the radioligand, but also by others of which the most important is the arterial plasma input of the radioligand. The latter is influenced by peripheral metabolism and by plasma free fraction. The propensities of the radioligand to bind in brain and/or periphery, either specifically and/or nonspecifically, are also other important factors.

Fig. (10).

Representation of the structure of the BBB showing pathways into brain and major efflux transporters.

Fig. (11).

Representation of the structure of the phospholipid bilayer at the BBB.

Methods for predicting the ability of a radioligand to enter brain from knowledge of molecular structure alone are keenly sought in order to avoid wasted efforts on candidate radioligand synthesis and evaluation. Developers of drugs for CNS disorders have a similar need for methods to predict BBB penetration. Hence, numerous studies have attempted to identify molecular parameters that may be key to attaining a high equilibrium ratio of the concentration of a compound in brain to that in plasma, which is termed BB or K_p, and often discussed as logBB. BB values have been readily obtained in rodents. Generally, a K_p value of > 2, corresponding to logBB > 0.3, has been sought in CNS drug development. Before considering the influence of efflux transporters, it is valuable to consider the many possible molecular factors that may influence the ability of a compound to traverse the BBB by passive diffusion alone.

Parameters that have been considered for predicting BB values are numerous but often related to compound size (MWt), lipophilicity indexes (logD^7.4 or logP, computed or measured), charge (pK_a), molecular volume (V_x), shape (cross-sectional area or principal axes ratio), polarity [e.g., topological polar surface area (TPSA)], and hydrogen bonding capability [hydrogen bond donors (HBDs), hydrogen bond acceptors (HBAs), and/or total heteroatom count]. Some of these parameters are clearly inter-dependent, for example MWt and V_x, or total heteroatom count and TPSA. A few of these parameters are readily calculated from simple 2-dimensional representations of molecular structure plus, in some cases, the use of readily available software packages (e.g., for calculating lipophilicity or TPSA). A myriad of other parameters has also been considered for ‘in silico’ prediction of compound ability to penetrate the BBB, but the estimation of these parameters is computationally more demanding.

Several ‘rules or thumb’ have been proposed for predicting compound penetration of the BBB on the basis of the more simply acquired molecular data. Examples are specified at the foot of Table 4. These rules highlight the importance of molecular weight, lipophilicity, absence of charge and low HBD capacity. Table 4 shows the maximal uptakes of 20 PET radioligands in rhesus monkey brain and human brain as SUV, as measured in our laboratory, along with many of their properties that may be used to predict BBB penetration by the listed simple ‘rules of thumb’. Inspection of this dataset shows that the majority of radioligands comply with the following criteria:

• MWt < 500 Da

• clogD^7.4 > 1 and < 5

• logP < 5

• pK_a of most basic site < 9.5

• HBDs < 3

• TPSA < 90 Ų

These criteria quite closely correspond to those that were concluded to be acceptable in another recent survey of 77 CNS PET radioligands [175]. An understanding of why such criteria show importance may be gained from numerous experimental studies that are exemplified as follows.

The criterion for MWt < 500 Da reflects the difficulty for large molecules to pass through the tight junctions of the BBB [177]. In this regard, Gleeson [174] reported on logBB for 3,059 diverse molecules from a GlaxoSmithKline database. LogBB declined almost linearly with MWt. Molecules with MWt < 300 Da had a mean BB value of 2.2 compared to 0.1 for molecules with MWt > 700 Da.

The criteria for logP, pK_a, and logD^7.4 reflect the inability of charged and hydrophilic molecules to readily undergo passive diffusion across the BBB. LogP values represent the partition coefficients for neutral compounds between an organic phase, n-octanol, and water. Ligand uptake into brain from plasma is expected to increase as logP increases. However, binding to blood proteins will also be expected to increase, eventually restricting entry into brain from plasma. Notionally, therefore, moderate lipophilicity is desirable for adequate brain entry. A preponderance of CNS drugs and brain PET radioligands either do not ionize or are basic amines. Highly basic compounds with pK_a > 9.5 will be almost fully protonated at the pH of blood (7.4) and mostly incapable of crossing the BBB. LogD^7.4 represents the distribution of the fraction of uncharged species between n-octanol and pH 7.4 buffer. For strong and moderately strong bases, logD^7.4 values are generally much lower than logP values, as may be seen in Table 4. Because charged species are well excluded from entering brain from plasma (pH 7.4), logD^7.4 is preferable to logP for assessing the influence of radioligand lipophilicity on brain uptake. A logD^7.4 value >1, showing preference of a compound for organic phase over aqueous phase, may be rationalized as clearly desirable for easy brain entry.

Considering only the lipophilicity parameter logP, Waterhouse [178] observed an approximately parabolic relationship between logP and rat whole brain uptake for a set of 43 candidate and 7 effective PET radioligands, with a peak in the range 2–3 for logP. However, whether such a parabolic relationship holds generally is perhaps uncertain: plots of peak uptake in monkey brain versus computed logP or measured logD^7.4 for the 20 radioligands listed in Table 4 do not show a similar relationship, only tendencies for peak brain uptake to decrease slightly with increasing lipophilicity (Fig. 12). Peak brain uptakes are however strongly correlated between monkey and human (Fig. 13).

Fig. (12).

Plot of peak radioactivity concentration in rhesus monkey brain (SUV) versus radioligand lipophilicity (measured logD) for the 20 radioligands listed in Table 4. Line represents linear fit (r² = 0.19).

Fig. (13).

Relationship of peak human brain uptake to peak monkey brain uptake (SUV) for 14 radioligands in Table 4. Line represents linear fit (r² = 0.74).

Norinder and Haeberlein [172] considered both MWt and logP parameters and found the following simple but strong relationship (r² = 0.89) with logBB for a set of 20 compounds:

$$\text{logBB}=-0.00116\text{MWt}+0.272\text{logP}-0.088$$ Equation 21

This equation also predicts positive logBB for 19 of the 20 the brain-penetrant radioligands in Table 4, with [¹¹C]CUMI-101 as the only exception.

The listed criteria for HBDs and TPSA reflect the recognized detrimental effect of high hydrogen bonding capacity on brain penetration [173, 179]. The high energy cost of compound desolvation is considered to be the reason why a high HBD number disfavors ready passive diffusion across the BBB. TPSA often strongly reflects the total number of HBDs, and is rapidly computable from two-dimensional structure with commercial software [180], such as ChemDraw. Clark [181] found the following relationship (r² = 0.79) between logBB, and the two parameters clogP and PSA (polar molecular surface area, which may be regarded as similar to TPSA) for 55 compounds:

$$\text{logBB}=0.152\text{clogP}-0.0148\text{PSA}+0.139$$ Equation 22

Moreover, Rishton et al. [182] found a very similar strong relationship (r² = 0.79) for 207 compounds:

$$\text{logBB}=0.155\text{clogP}-0.0100\text{PSA}+0.164$$ Equation 23

Decreasing the number of HBDs and/or decreasing TPSA have been found effective strategies for enhancing compound brain uptake in several studies [173]. Consistent with the criteria for lack of charge, lack of HBDs and hydrophobicity, certain polar structural features have been observed as detrimental to successful brain entry and include sulfonamido, sulfone, tetrazole, carboxylic acid or N-oxide groups [173].

Strenuous efforts have been made to move beyond simple rules-of-thumb to derive strong computational methods for predicting compound brain penetration. For example, Gerebtzoff and Seelig [183] closely studied the effects of drug charge and molecular shape on brain entry. For a set of 43 compounds, they computed molecular cross sectional areas (A_D) for the calculated conformation with strongest amphiphilicity, which they considered to be the membrane-binding conformation. They concluded that a ligand cross-sectional area of less than 70 Ų is required to allow brain entry. Use of this criterion gave 83% success in predicting brain penetration for a separate set of 42 compounds. However, lack of availability of the software for the computational procedures restricts wider testing of this promising but computationally challenging method.

Adenot and Lahana [184] investigated several models for their ability to predict BBB penetration for a set of 1,595 drugs. Overall, 67 variables were considered. This study exposed some simple yet powerful approaches to predicting BBB penetration, of which two are especially worthy of mention. Thus, by reference to Lipinski’s rules [170], Adenot and Lahana [184] derived a composite parameter C_diff based on five parameters (1–5), namely MWt (1), clogP (2), HBDs (3), HBAs (4), and TPSA (5), respectively. C_diff was defined as a pseudo-continuous variable ranging from 0 to 1 by the following equation:

$$C_{\text{diff}}={\sum }_{i=1,5}S(i)/5^2$$ Equation 24

in which S(i) is a scaling function for the ith parameter. Each function was characterized by its respective slope a_i and lower and upper thresholds, t_i and t_i′ respectively, with the following values: a₁ = −6.7 × 10⁻³, t₁ = 0, t₁′ = 800; a₂ = 0.4, t₂ = −2.5, t₂′ = 5; a₃ = −1, t₃ = 0, t₃′ = 5; a₄ = −0.5, t₄ = 0, t₄′ = 10; a₅ = −0.02, t₅ = 0, t₅′ = 250. C_diff above 0.4 correctly classified 95% of brain-penetrant compounds and 93% of non-penetrant compounds among the large training set. Similar performance was achieved on an external set of 82 compounds. An attractive feature of C_diff is its simple computation from 2-dimensional molecular structure.

Among the multitude of variables examined by Adenot and Lahana [184], a remarkably high performer was found to be the total number of heteroatoms (defined as O, S, N, P, and halogen), which correctly predicted 94% of brain-penetrant compounds and 72% of non-penetrant compounds, when the criterion for brain entry was set at less than 9 heteroatoms. Thus, simple heteroatom-counting performed almost as well as any of the numerous computationally more demanding methods for predicting BBB penetration. All radio-ligands in Table 4 meet this simple requirement.

The parameter logBB, although frequently used as an index of brain penetration, really represents the distribution of compound between plasma and brain where the compound is able to diffuse passively across the BBB in either direction [179, 185]. This parameter is strongly criticized as an index of compound brain penetration [186]. For example, logBB takes no account of ligand binding to proteins in plasma or to tissue in brain, whereas it is generally considered that only free ligand in plasma is available for brain entry. The rate constant K₁ (Fig. 3) more truly reflects the degree to which a compound is able to diffuse passively across the BBB.

Guo et al. [99] used an in-silico method to predict K₁ (mL/cm³/min) from perfusion (f; mL/cm³/min), compound permeability (P′; cm/min), and capillary surface area (S; cm²/cm³), as follows:

$$K_1=f\times (1-{\mathrm{e}}^{-{\mathrm{P}}^{\prime }S/f})$$ Equation 25

Perfusion and capillary surface area are known physiological parameters. P′ was calculated from computed lipophilicity (clogD^7.4) and molecular volume (V_x), according to the relationship:

$$\text{log}{\mathrm{P}}^{\prime }=0.121{({\text{clogD}}^{7.4}-2.298)}^2-2.544\text{log}({V_{\mathrm{x}}}^{1/3})-2.525$$ Equation 26

The coefficients of this parabolic relationship between P′ and clogD were derived empirically from 30 putative non-efflux transporter (P-gp) substrates. The parabola has an optimum at a clogD^7.4 value of 2.3, and shows a quite strong correlation with the empirical data (r² = 0.58). The Pearson coefficient (r) for the correlation between predicted K₁ and that measured in pig with PET was nonetheless quite low at 0.402. Three of these compounds have very low brain penetration and may be considered as outliers (perhaps due to efflux transporter action). Curve-fitting of the dataset excluding these 3 outliers shows a maximum for K₁ at a clogD^7.4 value of about 2 (Fig. 14).

Fig. (14).

Plots of predicted and measured K₁ versus lipophilicity (clogD) for 15 ligands in pig. The blue and red lines are Gaussian fits to the predicted and measured data, respectively. (Source data: Guo et al. [99]).

In view of the Guo et al. study [99], and of the earlier mentioned studies, candidate PET radioligands should probably be sought with clogD^7.4 in the range 1 to 4, and optimally about 2.2, for the purpose of achieving adequate brain penetration. In addition, as listed previously, MWt should preferably be < 500 Da, the pK_a of the most basic site < 9.5, HBDs < 3, TPSA < 90 Ų_, and heteroatom count < 9.

Finally, it should be noted that commercial software is now available to predict compound logBB, such as that from ACD/Labs. This program correctly classifies 92% of the large dataset of Adenot and Lahana [184], as either brain-penetrant or non-penetrant [187], after removing compounds that are putative P-gp substrates from the dataset.

9. EFFLUX TRANSPORTER ACTION

The preceding discussion on PET radioligand penetration of the BBB does not take into account the influence of efflux transporters. Brain efflux transporters are responsible for the inability of a high proportion of small molecules, including experimental therapeutics, to access brain. Of primary consideration and seemingly most relevant to PET radioligand development are transporters of the adenosine triphosphate binding cassette (ABC), which has 48 identified members. These include P-glycoprotein (P-gp; MDR1/mdr-1a; ABCB1), breast cancer resistant protein (BCRP; ABCG2), and multidrug resistance associated protein (MRP1; ABCC1) [188]. These are each expressed on the luminal (blood facing) side of the BBB (Fig. 10).

P-gp is less abundant than BCRP in human brain [179], but has been by far the most studied. Considerable species differences in the expression levels of these two transporters have been noted [179]. For example, the order of brain expression of P-gp is mouse > human > cynomolgus monkey, and for BCRP, cynomolgus monkey > human > mouse. Moreover, there exist considerable species differences in the sequence homologies of P-gp. Thus, guinea pig, rat, mouse, dog, rhesus monkey and chimpanzee share 82, 85, 87, 87, 93, and 97% homology with human P-gp, respectively [179]. Given these variations in P-gp expression level and in sequence, compounds may show considerable species differences in P-gp substrate behavior [189, 190]. Thus, Xia et al. [189] found appreciable differences in the affinities of 21 substrates for P-gp between beagle dog, rhesus monkey and human. Some PET radioligands show adequate brain uptakes (> 2 SUV) in human subjects, but markedly lower brain uptakes in rodents, perhaps suggesting a role for efflux transporters in rodents.

For many radioligands, it has been noted that genetic or pharmacological knockout of either P-gp (Table 5) or BCRP (Table 6) results in globally increased brain uptake of radioactivity in vivo. However, conclusions from such studies that a particular PET radioligand is indeed a substrate for P-gp or BCRP must be treated with some caution, because of other possible explanations. Thus, increases in the uptakes of PET radioligands into brain following pharmacological inhibition of P-gp have been ascribed to increases in radioligand f_P, in at least two cases, namely [¹¹C]RWAY in monkey [198], and [¹⁸F]MPPF in baboon [194]. In many other studies, full compartmental modeling with correction for radioligand f_P has been impractical or has not been applied.

Table 5.

Reports of P-gp transporter sensitivity for some PET radioligands across species, based on in vitro assays and on in vivo, mainly PET, studies.

Radioligand	Imaging target	In vivo species	In vitro species	Substrate (Yes/No)	Reference
[11C]Carazolol	β-AR	M (g)		Yes	[191]
[11C]Carazolol	β-AR	R (pc)		Yes	[191]
[18F]S-1′-fluorocarazolol	β-AR	M (g)		Yes	[191]
[18F]S-1′-fluorocarazolol	β-AR	R (pc)		Yes	[191]
[18F]FCWAY	5-HT1A		H (c)	Yes	[192]
[18F]FCWAY	5-HT1A	H (pt)		Yes	[192]
[18F]MPPF	5-HT1A	M (g)		Yes	[193]
[18F]MPPF	5-HT1A	M (g)		Yes	[194]
[18F]MPPF	5-HT1A	M (p)		Yes	[194]
[18F]MPPF	5-HT1A	R		Yes	[195]
[18F]MPPF	5-HT1A	R (pc)		Yes	[193]
[18F]MPPF	5-HT1A	R (pc)		Yes	[196]
[18F]MPPF	5-HT1A	NHP (pc)		No	[194]
[18F]MPPF	5-HT1A		H (c)	No	[197]
[11C]RWAY	5-HT1A	M		Yes	[198]
[11C]RWAY	5-HT1A	R		Yes	[198]
[11C]RWAY	5-HT2A	NHP		No	[154]
[11C]WAY	5-HT1A	R (pc)		Yes	[193]
[11C]WAY	5-HT1A		H (c)	No	[197]
[18F]Altanserin	5-HT2A	R (pc)		Yes	[190]
[18F]Altanserin	5-HT2A	R (pc)		Yes	[199]
[18F]Altanserin	5-HT2A	GP (pc)		Yes	[190]
[18F]Altanserin	5-HT2A	MP (pc)		Yes	[190]
[18F]Altanserin	5-HT2A	NHP (pc)		Yes	[190]
[18F]MH-MZ	5-HT2A	M (pc)		Yes	[200]
[11C]PIB	Aβ	M (pc)		No	[201]
[11C]PIB	Aβ		H (c)	No	[197]
[11C]Flumazenil	Bz	R (pt)		Yes	[202]
[11C]Flumazenil	Bz	M (pc)		Yes	[201]
[11C]Flumazenil	Bz		H (c)	No	[197]
[11C]MePPEP	CB1	M (g)		No	[203]
[11C]MePPEP	CB1	NHP (pd)		No	[164]
[18F]Fallypride	D2/D3	M (pc)		Yes	[204]
[18F]Fallypride	D2/D3	M (g)		Yes	[204]
[18F]Fallypride	D2/D3	R (pc)		Yes	[204]
[18F]Fallypride	D2/D3		H (c)	No	[197]
[11C]Raclopride	D2/D3	M (pc)		No	[201]
[11C]Raclopride	D2/D3		H (c)	No	[197]
[11C]Befloxatone	MAO-A	R (c)		No	[205]
[11C]Befloxatone	MAO-A		H (c)	No	[197]
[11C]GR205171	NK1	M (g)		Yes	[206]
[11C]GR205171	NK1	R (pc)		Yes	[190]
[11C]GR205171	NK1	GP (pc)		Yes	[190]
[11C]GR205171	NK1	NHP (pc)		Yes	[190]
[11C]PK11195	TSPO	M (pc)		No	[201]
[11C]PK11195	TSPO		H (c)	No	[197]

M = Mouse, R = Rat, GP = Guinea pig; MP = mini-pig; NHP = non-human primate; H = Human, c = cell-based assay; g = efflux transporter knockout; pc = pharmacological block of efflux transporter with cyclosporin A; pd = pharmacological block of efflux transporter with DCPQ; pt = pharmacological block of efflux transporter with tariquidar.

Table 6.

Reports of BCRP transporter sensitivity for some PET radioligands in various species, based on in vitro assays and on in vivo, mainly PET, studies.

Radioligand	Imaging target	In vivo species	In vitro species	Substrate (Yes/No)	Reference
[18F]MPPF	5-HT1A	M (g)		No	[194]
[18F]MPPF	5-HT1A		H (c)	No	[197]
[11C]WAY	5-HT1A		H (c)	No	[197]
[11C]PIB	Aβ		H (c)	No	[197]
[11C]Flumazenil	Bz		H (c)	No	[197]
[18F]Fallypride	D2/D3		H (c)	No	[197]
[11C]Raclopride	D2/D3		H (c)	No	[197]
[11C]Befloxatone	MAO-A	R (g)		No	[205]
[11C]Befloxatone	MAO-A		H (c)	Yes	[197]
[11C]PK11195	TSPO		H (c)	No	[197]

M = Mouse, R = Rat; H = Human.

c = cell-based assay; g = efflux transporter knockout.

The behavior of [¹¹C](R)-RWAY exemplifies how species differences in efflux transporter action may dramatically affect radioligand performance. This radioligand readily enters rhesus monkey and human brain; and has found application in studies requiring brain 5-HT_1A receptors to be measured in rhesus monkey [154, 155]. However, P-gp greatly prevents appreciable uptake of this radioligand into rodent brain [198]. As a general observation, species differences in efflux transporter action between NHPs and humans have seldom been observed, unlike those between rodents and primate.

Some radioligands that do show adequate uptake in human brain for target quantification are nonetheless possibly weak substrates for efflux transporters. The substrate behavior of some PET radioligands towards human efflux transporters has been tested with in vitro assays, making use of cells expressing human P-gp [197] (Tables 5 and 6). However, investigation in vivo is difficult because of a paucity of efflux transporter inhibitors that may be given effectively and safely to human subjects. Currently, tariquidar seems most promising for inhibiting P-gp in human subjects to test for PET radioligand substrate behavior [207]. In this regard, the 5-HT_1A receptor radioligand [¹⁸F]FCWAY has recently been identified as a weak substrate of P-gp in human subjects by using tariquidar to inhibit P-gp [192].

Considerable efforts have been made to identify structural features or molecular properties that may impart the presence or absence of P-gp substrate behavior [208, 209]. Gleeson [174] found that compound susceptibility to P-gp efflux increased markedly over the MWt range 300–700 for 1975 diverse molecules. Among this large set of compounds ionization was found to play a minor role in extent of P-gp efflux. Compounds with logP between 3 and 5 were more prone to P-gp efflux than compounds with lower or higher logP. The effect of lipophilicity was independent of MWt. Earlier in this article, it was argued that moderate lipophilicity is desirable for facile radioligand penetration of the BBB. Thus, it is desirable for a radioligand to have moderate lipophilicity without efflux transporter substrate behavior, and to identify molecular criteria that might achieve this aim.

Seelig et al. [210, 211] proposed that P-gp substrate behavior was conferred if a molecule contained two or three HBAs separated by 2.5 or 4.6 Å. Hitchcock [212] reviewed the structural features that alter compound P-gp efflux susceptibility and suggested that susceptibility is minimized in compounds that have < 2 HBDs, and TPSA < 90 Ų, with a preferred value of < 70 Ų. From an analysis of 2000 compounds at Eli Lilly, Desai et al. [213] found that compounds with TPSA < 60 Ų and with pK_a calculated to be < 8 exhibited a much lower likelihood of being a P-gp substrate (6%) than those with parameters outside these limits (75%). From this dataset, a QSAR model was developed that predicted both P-gp substrates and non-substrates with 80% success. Didziapetris et al. [214] proposed that molecules with MWt < 400 Da, with < 7 nitrogen plus oxygen atom HBAs, and having a basic pK_a of < 8 are unlikely to be P-gp substrates. This became dubbed the ‘rule of 4’ and compliance with this rule appears to be a useful strategy for avoiding efflux transporter substrate behavior [182].

In summary, the suggested criteria for MWt, TPSA, pK_a and HBD for avoiding liability to efflux transporter substrate behavior are within the limits that were earlier considered desirable for facile penetration of the BBB by passive diffusion. However, it is clear that whether a particular PET radioligand will be a substrate for efflux transporters in any particular species is still not easily predictable. Currently, data on the efflux transporter susceptibility of PET radioligands for brain imaging are sparse, and accumulation of more data are needed to have a deeper understanding of this area.

10. ADEQUATE AND MEASUREABLE FREE FRACTION IN PLASMA

The free fraction of a radioligand in plasma (f_P) is an important parameter because it is assumed that only radioligand that is free in plasma may cross the BBB for binding in brain. f_p may vary appreciably between subjects and between different experimental scenarios, such as between baseline and target blocking conditions. Ideally, comparisons of V_T values should be corrected for significant differences in f_P under compared conditions i.e., it is generally preferable and more rigorous to compare V_T/f_p values than V_T values. Equilibrium dialysis of the non-radioactive tracer is regarded as the ‘gold standard’ method for f_p determination [215, 216], but is relatively slow and inapplicable to PET experiment settings. For PET experiments, f_P is usually measured by a rapid ultrafiltration technique with the radioligand preparation, using blood sampled from the subject of study [217]. Where f_P measurements are used, they need to be accurate and precise. Experimentally, this is more easily achieved when the f_p value itself is substantial.

In general, f_p tends to decrease with increasing lipophilicity, but not linearly so. Zoghbi et al. [176] generated the parameter ^mlogD_pr/pl which is the measured logarithm of the distribution of radioligand between plasma protein and plasma, where

$${\mathrm{D}}_{\text{pr}/\text{pl}}=(1-f\mathrm{p})/f\mathrm{p}$$ Equation 27

For a set PET radioligands ^mlogD_pr/pl was linearly related to measured logD value by the following equation for human:

$${}^m\mathrm{l}{\text{ogD}}_{\text{pr}/\text{pl}}=0.733{}^m\mathrm{l}\text{ogD}-0.651(r^2=0.738)$$ Equation 28

and for monkey:

$${}^m\mathrm{l}{\text{ogD}}_{\text{pr}/\text{pl}}=0.780{}^m\mathrm{l}\text{ogD}-1.15(r^2=0.832)$$ Equation 29

The strong dependency of f_P on compound lipophilicity has also been noted in other reports [174, 218]. Generally, the following trend in f_P has been observed: acids > neutrals > zwitterions > bases [178]. From equations 28 and 29, a radioligand that has a logD of about 2.2 may be considered appropriate for achieving acceptable passage of the BBB, and would be expected to have f_P values of about 0.10 and 0.21 in human and monkey, respectively. These values would be easily measurable with accuracy and precision.

11. LOW NONSPECIFIC BINDING (OR LOW NON-DISPLACEABLE BINDING)

All radioligands for PET imaging of brain proteins display accompanying nonspecific binding. Excessive nonspecific binding masks any specific binding and so can render a radioligand of little or no utility. An overall objective in developing a PET radiotracer for brain imaging is therefore to achieve a high signal-to-noise ratio, as represented by a high ratio of V_T to V_NS, where V_NS is V_ND minus V_F (Equation 11). For a radioligand which crosses the BBB by free passive diffusion only, V_F is simply 1/f_ND. Hence, in such cases, it follows that V_NS/f_P is proportional to V_ND/f_P according to Equation 30:

$$V_{\text{NS}}/f_{\mathrm{P}}=V_{\text{ND}}/f_{\mathrm{P}}-1$$ Equation 30

Recalling Equation 12 for radioligands that cross the BBB by passive diffusion only, nonspecific binding (V_ND) theoretically equates to the ratio of plasma free fraction (f_P) to brain free fraction (f_ND). V_ND measurements from PET and f_p and f_ND data from equilibrium dialysis provide experimental support for the validity of this relationship in pig [219]. In this study of 36 PET radioligands, V_ND varied from 0.23 to 29.7. Radioligands with unexpectedly low V_ND values were considered likely to be efflux transporter substrates. Moreover, in another study, f_P values have been found to be highly correlated across three species (rat, pig and human (r² > 0.81, n = 21), so indicating relatively uniform microenvironments for ligand binding in different species of brain tissue [220]. Brain has an overall different composition to plasma. The lipid content of brain (11% by volume) is 20-fold higher than that of plasma whereas plasma has twice as much protein as brain [221]. Therefore, it is unsurprising that f_ND is often very different to f_p for particular radioligands.

Because of the high lipid content of brain, it is generally expected that nonspecific binding in brain will increase with radioligand lipophilicity. Some authors have attempted to relate surrogate parameters of non-specific binding and logD^7.4 for PET radioligands. For example, Andersson et al. [222] reported a weak linear correlation (r² = 0.50) between radioactivity in cerebellum at 60 min, assumed to be nonspecific and non-displaceable binding, and clogD^7.4 over a narrow range for 8 experimental 5-HT_1B receptor radioligands in cynomolgus monkey. However, possibly more meaningful relationships between V_NS or V_ND and clogD^7.4 were not tested. Assmus et al. [223] found a good correlation (r² = 0.88) between logV_NS/f_P values determined with PET in human and estimates of log partition coefficients between brain tissue and water at pH 7.4 (logD_brain). These logD_brain estimates were obtained in a lipid membrane binding assay for a small set (n = 6) of well-known PET radioligands that were considered non-efflux transporter substrates. Assmus et al. also found only a very poor correlation between radioligand logD^7.4 and the log of nonspecific binding (logNSB) estimated from in vitro autoradiography. However, the estimates of logNSB were not strictly controlled for factors affecting nonspecific binding in autoradiography, such as the ligand concentration to which tissue was exposed, which ranged widely (0.08–23.5 nM). Correlations were much stronger for logNSB versus logD_brain. Gleeson [174] found that compound binding to brain tissue had a similar dependence on lipophilicity as plasma protein binding in an analysis of over 986 diverse compounds from a GlaxoSmithKline database. This correlation was largely independent of ligand molecular weight. Also in this study, binding to brain tissue was not as greatly affected by ligand ionization state as was plasma protein binding.

The identification of possible mechanisms underpinning nonspecific binding beyond hydrophobic interactions has received little attention. Rosso et al. [224] applied quantum mechanical modeling to show a correlation between estimates of the energies of drug molecule-lipid molecule interactions with in vivo nonspecific binding as estimated by PET for a small set of radioligands. The P-gp efflux radiotracer [¹¹C]loperamide is moderately basic (pK_a = 7.3) and has been shown to accumulate in acidic lysosomes (pH ~ 5.5) due to entrapment of its protonated form [225]. Lysosomes appear to have quite high capacity for such entrapment [225], and therefore conceivably similar ionic entrapment might contribute to the ‘nonspecific’ binding of radioligands or their radiometabolites with similar pK_a.

Nonspecific binding (V_ND) is quite readily determined in animal PET experiments as being equivalent to the V_T in experiments where radioligand specific binding has been completely prevented pharmacologically (as earlier illustrated for [¹⁸F]FIMX [107] (Fig. 6). Moreover, a ‘Lassen plot’, which in its simplest and original form is a plot of brain regional V_T in a baseline experiment (V_ND + V_S) versus the regional reductions in V_T in a partial target block experiment, provides V_ND as the Y-axis intercept [226]. This type of plot, and its more recent variations [227], are useful even in the absence of a target-free reference region and can enable measurement of drug occupancy at the target in response to variation of administered dose. Determinations of V_ND in human subjects are not always accessible in this manner, because of the absence of a known blocking agent with adequate brain penetration and safety. An interesting approach to overcome this obstacle is a modified Lassen plot in which the brain regional expression of target mRNA is used as a surrogate for V_S on the Y-axis. This approach worked satisfactorily for determining the human V_ND for [¹⁸F]FIMX [228]. It may be noted that because of gross interspecies similarities in brain compositions, V_ND is not expected to vary greatly between species, and particularly between primates. Brain regional variations in V_ND also appear rare.

12. LACK OF ACCUMULATION OF RADIOME-TABOLITES IN BRAIN OR SKULL

A PET camera is unable to assign detected radioactivity to particular molecular sources, and therefore is simply unable to distinguish parent radioligand from radiometabolites. For brain imaging with PET, the radioactivity in brain is ideally ascribable to parent radioligand alone [19, 100]. If this is not the case, then radiometabolites in brain may distort or confound attempts at quantifying radioligand-target protein interactions, for example, by contributing erroneously to radiotracer V_T and V_ND estimates derived from compartmental models. The time-stability of these estimates may also be imperiled by brain radiometabolites. Where radiometabolites also show specific binding, V_s can also be affected. Radiometabolites in brain also decrease the ‘signal-to-noise’ ratios, i.e. the V_S/V_ND value. Only in relatively few cases, such as for [¹⁸F]altanserin, where either a constant infusion paradigm [229] or a reference tissue model [230] can be reliably applied, are brain radiometabolites of less concern. In these cases, it has to be assumed that radiometabolites contribute solely to nonspecific binding in a uniform manner across brain.

The vast majority of PET radioligands do undergo rapid and extensive metabolism over the time-spans of PET scanning sessions. This metabolism may occur in periphery and/or brain [100]. For reversibly binding PET radioligands for brain imaging, such metabolism is ideally confined to the periphery. Some radioligands produce radiometabolites in the periphery that are more lipophilic than the parent radioligand, and this can be especially problematic as the radiometabolites may have enhanced entry into brain. An example comes from a DAT radioligand, [O-methyl-¹¹C]β-CIT, which produces the more lipophilic and brain-penetrant [O-methyl-¹¹C]nor-β-CIT as a radiometabolite [231]. For most radioligands, however, peripheral metabolism results in radiometabolites showing lower lipophilicity than the parent radioligand. In such cases, the blood-brain barrier may often serve beneficially to preclude extensive entry of circulating radiometabolites into brain. This is an important consideration in the design of any brain radioligand, particularly with respect to selecting the molecular position for the label. As a caution, although peripheral radiometabolites that are less lipophilic than the parent radioligand are generally considered to have much less ability to cross the BBB, this cannot always be assumed. Thus, in the example of [¹⁸F]altanserin, a significant radiometabolite is the less lipophilic alcohol, [¹⁸F]altanserol, generated from reduction of its ketone moiety, which does enter brain [232].

An often favored position for a carbon-11 label is in an aryl O-methyl group, as seen in [¹¹C]raclopride, [¹¹C]PBR28, [¹¹C](R)-rolipram, [¹¹C]RX-1, [¹¹C]Me-PPEP, [¹¹C]CUMI-101 and [¹¹C]RWAY (Table 4). Such groups are often susceptible to demethylation in the periphery to generate [¹¹C]formaldehyde, which is further oxidized to [¹¹C]carbon dioxide with negligible radioactive contamination of brain. A radioligand with a carbon-11 label in a secondary N-methyl-amino position may be metabolized similarly and therefore just as benignly. However, a radioligand with a carbon-11 label in a tertiary N,N-dimethylamino group has a risk of being cleaved by monoamine oxidase within the brain to release [¹¹C]dimethylamine, which then becomes protonated and unable to leave brain [233], or of being metabolized to the radioactive secondary methylamine, which may well enter brain and in some cases show some specific binding to the target or other proteins. A notable exception is the SERT radioligand, [¹¹C]MAD-AM, which performs acceptably well with the carbon-11 label in its dimethylamino position [234].

Tolyl compounds are known to be susceptible to oxidative metabolism to the corresponding aryl hydroxymethyl compounds and then through to the aryl carboxylic acid. A radioligand having a tolyl group has a risk of producing the radioactive hydroxymethyl compound and thenceforth the radioactive aryl carboxylic acid. An example is [¹¹C]PE2I, for which the radioactive hydroxymethyl metabolite is brain-penetrant in rat [235], and also possibly in human [236]. Although aryl methyl groups are amenable to labeling with carbon-11, caution is clearly warranted with regard to considering an aryl methyl group as a prospective site for a carbon-11 label.

Labeling with carbon-11 in an aryl O-methyl position is not guaranteed to avoid issues from radiometabolites, because metabolic transformations at other sites in such a radioligand may be significant and result in metabolites that retain the radiolabel. The 5-HT_1A receptor radioligand, [O-methyl-¹¹C]WAY-100635, provides a classic example. In primates, including human, this radioligand is predominantly metabolized by amide hydrolysis to give a brain-penetrant and pharmacologically active amine radiometabolite, known as [¹¹C]WAY-100634 (Fig. 15) [237]. Although [O-methyl-¹¹C]WAY-100635 provided the first images of 5-HT_1A receptor in human brain [121], the same ligand when labeled in the amide carbonyl position gave images with much lower nonspecific binding in cerebellum and hence higher contrast for receptor-rich regions [122]. For the carbonyl-labeled ligand, amide hydrolysis results in [carbonyl-¹¹C]cyclohexane-carboxylic acid and not [O-methyl-¹¹C]WAY-100634 as the radiometabolite [238]. The radioactive acid has poor ability to enter and accumulate in brain, and consequently nonspecific binding is decreased. Labeling on the carbonyl side of a metabolically vulnerable amide bond can therefore be an effective strategy to avoid extensive brain-penetrant radiometabolites. A similar strategy can be effective for metabolically vulnerable esters, as exemplified by the 5-HT₄ receptor radioligands, [¹¹C]RX-1 and [¹⁸F]RX-2 [152].

Fig. (15).

Metabolism of [O-methyl-¹¹C]WAY-100635 and [carbonyl-¹¹C]WAY-100635 in primates. Labeling in the carbonyl position avoids generating the brain-penetrant radioactive amine [O-methyl-¹¹C]WAY-100634.

A useful test of whether radiometabolites contaminate the PET signal seen in brain is whether placement of a radiolabel in another molecular position of the same ligand results in the same signal. Identical signals provide some assurance that radiometabolites are not an issue, whereas different uptakes indicate brain contamination by radiometabolite for at least one version of the radioligand. Thus, [¹¹C]flumazenil and [¹⁸F]flumazenil have highly comparable human brain kinetics, thereby indicating that brain-penetrant radiometabolites are not an issue for either of these radioligands [239]. ¹¹C-Labeled methyl esters are readily prepared and have been explored as prospective radioligands. An example is the DAT radioligand [¹¹C]LBT-999. Aside from expected hydrolysis by widespread esterases, [¹¹C]LBT-999 is susceptible to metabolism by cytochrome P450 enzymes through tolyl group oxidation, N-dealkylation and tropane ring dehydrogenation [240]. LBT-999 labeled with fluorine-18 in its N-fluorobut-2-enyl group gave very different signal to [¹¹C]LBT-999 in baboon brain [241]. Neither radioligand was advanced to evaluation in human in anticipation of metabolism likely confounding quantification of radioligand-DAT interaction. In the studies of LBT-999, incubations in rat and liver microsomes with subsequent LC-MS-MS, and also radio-HPLC analysis of radiometabolites from excised rat brain, and of rat and human plasma following radioligand administration, were some of the techniques used to elucidate metabolic pathways.

Whereas radioanalytical methods can be applied to identify and measure radiometabolites in rodent brain following radioligand administration, this is clearly not so for human brain. Inspection of the stability of a quantitative outcome measurement, such as V_T, with respect to the duration of imaging data used to make the measurement may give some indication of whether radiometabolites enter brain or not. Highly unstable V_T values are indicative of radiometabolite contamination in brain, and stable values indicative of a lack of significant contamination. Generally, V_T values that stabilize within 10% of a terminal value are considered acceptable. Many such examples have been reported for stable V_T in monkey and in human. One radioligand showing exceptionally stable V_T in human is [¹⁸F]FPEB [242].

The 5-HT_1A receptor radioligand [O-methyl-¹¹C](R)-RWAY [127] was designed as a more easily labeled alternative to [carbonyl-¹¹C]WAY-100635, with an expected lower susceptibility to amide hydrolysis as a result of having a reversed amide bond direction. This radioligand was found to perform well in rhesus monkey, as judged by V_T stability [154]. However, this radioligand failed to give adequately stable V_T in human, likely because of increased generation and brain ingress of a moderately lipophilic radiometabolite that appears to correspond to that seen in monkey [156]. Hence, effective application of this radioligand has been confined to studies in monkey [155]. Findings with this radioligand affirm that even subtle species differences in extent or nature of metabolism may be critical to radioligand performance. In detailed studies, species differences in metabolic pathways and the rates of radiometabolite generation are invariably observed [232, 243, 244]. Generally, metabolism tends to be faster in experimental animals, especially rodent and pig, than in human subjects. Possible adverse metabolic effects therefore pose uncertainty when advancing an experimental radioligand to human evaluation following encouraging findings in animals.

Labeling with carbon-11 in the N-methyl position of a secondary amide has also proven to be effective in several instances, as for example in [¹¹C](R)-PK11195, [¹¹C]ER176, and [¹¹C]NOP-1A (Table 4). However, radioligands with a carbon-11 label in a tertiary di-methylamido group would be potentially susceptible to demethylation to give a brain-penetrant radiometabolite, as exemplified by the P-gp efflux transporter radiotracer [¹¹C]loperamide, which demethylates to [¹¹C]desmethyl loperamide ([¹¹C]dLop) [245]. [¹¹C]dLop is in fact preferred as a radiotracer for the study of P-gp efflux transporter function, because of its less troublesome radiometabolite profile [246].

¹⁸F-Labeled ligands raise their own specific concerns, of which susceptibility to radiodefluorination is paramount. Peripheral radiodefluorination of a radioligand will produce [¹⁸F]fluoride ion in plasma which will then be taken up in bone, including skull. High radioactivity in skull compromises accurate measurement of radioactivity in brain regions near the skull, such as in neocortex, because of ‘spill-over’ effects. Table 7 summarizes observations on the susceptibility of ¹⁸F-labeled sub-structures to radiodefluorinate in human subjects. In general, ¹⁸F-fluorine atoms bound to carbons in homoaromatic groups (entries 1–6) and in substituted pyridinyl moieties (entries 7–13) show strong resistance to metabolic defluorination. [¹⁸F]SP203 (entry 14) shows extensive radiodefluorination in rat and monkey but not in human, although there is moderate accumulation of radioactivity in skull [166]. This skull radioactivity is considered to be radioactivity in red marrow rather than in bone itself, based on the whole body radioactivity distribution showing high uptake in bone structures with high red marrow content, such as lumbar vertebrae [259]. Thus, species differences in the susceptibility of radioligands to radiodefluorination can occur.

Table 7.

Examples of labeled sub-structures found in ¹⁸F-Labeled PET radioligands studied in humans, and whether noted to be susceptible to radiodefluorination.

Entry	Labeled sub-structure	Radioligand examples	Noted defluorination susceptibility	References
1		[18F]Flumazenil	No	[239]
2		[18F]Flumetamol	No	[247]
3		p-[18F]MPPF	No	[248]
4		[18F]FIMX	No	[228]
5		[18F]Altanserin	No	[249]
6		[18F]FPEB	No	[250]
7		2-[18F]F-A85380	No	[251] [252]
8		[18F]MK-6577	No, but some skull radioactivity uptake	[253]
9		[18F]AZAN	No	[254]
10		[18F]T807	Negligible	[255]
11		[18F](-)-Flubatine	Negligible	[256]
12		6-[18F]F-A85380	No	[257]
13		[18F]UCB-H	No	[258]
14		[18F]SP203	Low, but radioactivity uptake in red marrow	[259]
15		[18F]Fallypride	No	[260]
16		[18F]FMPEP-d2	Yes	[140]
17		[18F]FMENER-d2	Yes	[261]
18		[18F]SPA-RQ	Low, but some radioactivity uptake in red marrow	[262]
19		[18F]Fluoromethyl-McN5652	No	[263]
20		[18F]DPA-714	No but some skull radioactivity uptake - presumed to be in marrow	[264]
21		[18F]MNI-659	No but some skull radioactivity uptake - presumed to be in marrow	[265]
22		[18F]MNI-444	No, but low skull radioactivity up-take	[266]
23		[18F]MK-9470	No	[267]
24		[18F]FEOBV	No	[268]
25		[18F]FEPPA	No	[269]
26		[18F]THK-523	No, but low radioactivity uptake in skull	[270]
27		[18F]FESPA-RQ	Yes, but low	[271]
28		[18F]FEDAA1106	No	[272]
29		[18F]FACT	No, but low radioactivity uptake in skull	[273]
30		[18F]AV45	No	[274]
31		[18F]FP-(+)-DTBZ ([18F]AV-133)	No	[275]
32		Trans-[18F]FCWAY	Yes	[276]
33		[18F]Mefway	No	[277]
34		[18F]FBR	Negligible	[278]
35		[18F]FECNT	No	[167] [279]
36		[18F]JNJ42259152	No	[280]
37		[18F]CPFPX	No	[281]
38		[18F]FPCIT	No	[282]
39		[18F]LBT-999	Yes	[283]

Fluorine-18 present at the terminus of a straight saturated alkyl chain of two or more carbons appears quite resistant to defluorination in human subjects (entries 15, 19–28, 30, 31, 33). However, fluorine-18 at the terminus of the unsaturated carbon chain of [¹⁸F]LBT-999 is highly susceptible to defluorination (entry 39), as is fluorine-18 attached to the secondary carbon in [¹⁸F]FCWAY (entry 32). The severe radiodefluorination of [¹⁸F]FCWAY that leads to very high uptake of radioactivity in human skull can be countered by pre-inhibition of the enzyme identified as being responsible, namely CYP2E1; the required inhibition is simply achievable with a single oral dose of Antabuse (disulfiram) [284]. Because peripheral radiodefluorination is inhibited, the radioligand is more available for brain entry and therefore enhanced images are obtained in the absence of appreciable skull radioactivity. Inhibition of metabolic enzymes has wider applicability for enhancing radioligand performance. Thus, inhibition of cytochrome CYP1A2 by fluvoxamine has been suggested for reducing the metabolism of the A1 adenosine receptor PET radioligand, [¹⁸F]CPFPX (entry 37) [285]. Also inhibition of catechol O-methyl transferase with tolcapone has been proposed for avoiding a troublesome radiometabolite from the D₁ receptor agonist radioligand, (R)-[¹¹C]SKF 82957 [286]. A recently developed 5-HT_1A receptor radioligand, [¹⁸F]Mefway, overcomes the defluorination issue seen with trans-[¹⁸F]FCWAY by attaching fluorine-18 to a primary alkyl carbon (entry 33).

Radioligands labeled with fluorine-18 in aryl methoxy groups are well-known to be susceptible to radiodefluorination. Dideuteration of the methoxy group may reduce this susceptibility through an isotope effect, as for [¹⁸F]FMPEP-d₂, [¹⁸F]MENER-d₂, and [¹⁸F]SPA-RQ (entries 16–18). As dideuteration is easily achievable, this has become virtually standard practice for this mode of labeling.

Aside from radiodefluorination, ¹⁸F-labeled prosthetic groups in PET radioligands may present other issues. Thus, the N-[¹⁸F]2-fluoroethyl group in [¹⁸F]FECNT (entry 35) is susceptible to oxidative metabolism from liver cytochrome P-450 enzymes to produce [¹⁸F]2-fluoroacetaldehyde, which may metabolically interchange with [¹⁸F]2-fluoroethanol and [¹⁸F]fluoroacetate. These two-carbon radiometabolites are able to traverse the BBB to confound brain radioligand quantification [167]. Other candidate radioligands, having an N-[¹⁸F]2-fluoroethyl group, such as [¹⁸F]FAMV for imaging vesicular acetylcholine transporter, also have limited utility because they likely produce [¹⁸F]fluoroacetate as a brain-penetrant radiometabolite [287]. [¹⁸F]Fluoroacetate itself may also be defluorinated within the brain. By contrast [¹⁸F]FBR (entry 34), which has an [¹⁸F]fluoroacetamido group, shows little tendency to radiodefluorinate, as do the [¹⁸F]N-3-fluoropropyl groups in [¹⁸F]CPFPX (entry 37) and [¹⁸F]FPCIT (entry 38).

Finally, it should be noted that rapid peripheral metabolism of a radioligand to non brain-penetrant radiometabolites can be helpful rather than detrimental to radioligand performance, by decreasing radioligand in plasma and thereby free and nonspecifically bound radioligand in brain, which are all ideally in rapid equilibrium.

13. AMENABILITY TO ACCURATE QUANTIFICATION

Important questions with regard to the performance of a PET radioligand for brain imaging are not only the signal-to-noise ratio but also the utility of the radioligand for accurate quantification of the imaging target. The desirability of reliable and robust output measures, such as V_T, V_T/f_P, or BP_ND, was discussed above with regard to these measures being free of influences from radiometabolites in brain. As earlier mentioned, radioligands giving output measures that persistently stay within about 10% of the terminal value are usually regarded as adequately amenable to accurate quantification. The ability to derive stable measures is also dependent on the kinetics of the radioligand. Generally, PET radioligands intended to quantify brain proteins through reversible high-affinity binding should have sufficiently fast kinetics to show both accumulation and washout of radioactivity from the brain during the period of scanning [14]. Ideally, peak radioactivity in brain will occur relatively early so that the scanning session may be reasonably short and logistically acceptable to both subject and investigator. Data from the washout phase need to be adequate for accurate identifiability of kinetic parameters in the applied biomathematical model. For this purpose, it is typically desirable to acquire data covering more than 25% of the decline in radioactivity within brain.

14. AMENABILITY TO LABELING WITH ¹¹C OR ¹⁸F

Research on developing methods for labeling compounds with carbon-11 or fluorine-18 to generate PET radiotracers has intensified over the last four decades, and has now resulted in an impressive arsenal of methods that renders most small drug-like molecules amenable to labeling in at least one molecular position. This methodology has been frequently reviewed at varying depths and from various perspectives, for example, either generally [288, 289], or with focus on either carbon-1l [290–293] or fluorine-18 [294–300]. Only a brief survey of this methodology is given here from the narrow perspective of developing reversibly binding brain PET radioligands.

A primary requirement of any reversibly binding PET radioligand for imaging a protein target within the brain is for the carrier accompanying the radioligand to not cause appreciable occupancy of the available protein, and thereby invalidate the tracer assumptions underlying common quantification methods. In general, it is considered that any target occupancy should be kept below 5% [301]. Imaging targets are almost invariably present at very low densities (Tables 2 and 3), and therefore reversibly binding PET radioligands must be produced with low amounts of carrier that correspond to high specific activities. This is especially true for experiments in small animals [302, 303]. These specific activities are typically required to be greater than 40 GBq/μmol at time of administration to human, and even higher specific activities are always desirable.

The requirement for high specific activity imposes restrictions on the types of radiochemistry that can be used for preparing radioligands. Generally, reactions involving isotope exchange are ruled out because these will inevitably introduce carrier. Methods for producing the radionuclide, either carbon-11 or fluorine-18, must be no-carrier-added (NCA), and preferably very high yielding from modern medium-energy biomedical cyclotrons. Ideally, labeling procedures need to involve as few steps as possible, and be efficient and fast, to conserve activity as well as specific activity. The importance of the position of label for radioligand efficacy has already been stressed in this review. Hence, radiochemistry methods need to be available for labeling candidate PET radioligands at a position of choice. Carbon-11 and fluorine-18 pose different radiochemical challenges and possibilities, as summarized below.

14.1. Carbon-11

For carbon-11, the requirement for production at high yield and specific activity is well satisfied by the ¹⁴N(p,α)¹¹C reaction on nitrogen gas [304]. Several hundred GBq of carbon-11 can be prepared through this reaction on modern biomedical cyclotrons. Radioligands are produced with specific activities typically in the range 180–370 GBq/μmol [305, 306]. With rigorous precautions in production and labeling methodology, radioligands with impressively higher specific activities of up to about 5550 GBq/μmol may be obtained [305]. The chemical form obtained from the cyclotron irradiation may be [¹¹C]carbon dioxide [307] or [¹¹C]methane [308, 309], depending on whether the nitrogen gas target is doped with oxygen or hydrogen, respectively (Fig. 16). Radiochemistry aimed at preparing ¹¹C-labeled radioligands must begin with either of these two primary cyclotron-produced radioactive species, which may if required be readily interconverted on-line after cyclotron production [310, 311]. [¹¹C]Methane has no direct labeling utility, whereas [¹¹C]carbon dioxide has considerable reactivity that may be exploited to introduce a variety of labeled functional groups into radiotracers, including carboxyl, symmetrical and unsymmetrical urea groups, and carbamate groups [312]. Although alkyl and aryl carboxyl functions may be directly labeled with [¹¹C]carbon dioxide through reactions with for example Grignard reagents [312] or boronates [313], compounds with carboxyl groups are seldom effective PET radioligands for brain imaging because of their poor abilities to cross the BBB. Nonetheless, such ¹¹C-labeled carboxylic acids can generally be obtained rapidly in high yields, affording opportunities for onward conversion to other labeled compounds such as [¹¹C]amides [312, 313]. [¹¹C]N-Methylation has been achieved by direct [¹¹C]carboxylation of silanamines followed by reduction to prepare, for example, the D₁ receptor radioligand [¹¹C]SCH 23390 (Fig. 17) [314]. This method has not gained traction because of the general storage instability of the silanamines. In a recent adaptation, it has been shown that [¹¹C]carbon dioxide may be used to achieve [¹¹C]N-methylation reactions in two-steps [315]. This is an example of a recent drive towards more directly exploiting readily accessible [¹¹C]carbon dioxide in new labeling procedures for diverse functional groups [312], such as for labeling ureas [316] and carbamates [317] (Fig. 18). A major practical consideration in such direct uses of [¹¹C]carbon dioxide is to avoid carrier dilution by adventitious ingress of atmospheric carbon dioxide and reaction with precursor preceding or during the labeling reaction. Notwithstanding recent progress on advancing the utility of [¹¹C]carbon dioxide for direct labeling, primary cyclotron-produced ¹¹C-labeled precursors are still usually transformed into secondary labeling agents preceding any late-stage labeling reactions.

Fig. (16).

Cyclotron methods for preparing [¹¹C]carbon dioxide or [¹¹C]methane, and methods for their on-line intercon-versions.

Fig. (17).

Preparation of [N-methyl-¹¹C]amines via ¹¹C-carbonylation of amines.

Fig. (18).

Utility of [¹¹C]carbon dioxide for labeling ureas and carbamates.

By far the most widely applied secondary labeling agent is [¹¹C]methyl iodide [318]. This labeling agent has direct utility for attaching a [¹¹C]methyl group to carbon in aryl rings or alkyl chains, or to heteroatoms, such oxygen in phenols and carboxylic acids, sulfur in thiophenols, and nitrogen in amines and amides (Fig. 19). For several years the production of [¹¹C]methyl iodide was based on reduction of cyclotron-produced [¹¹C]carbon dioxide with lithium aluminum hydride followed by treatment of the adduct with hydroiodic acid [319]. This so-called ‘wet procedure’ is especially prone to carrier dilution from the reaction of LAH reagent with any ingress of atmospheric carbon dioxide [320–322]. Methods for producing [¹¹C]methyl iodide by direct high temperature iodination of cyclotron-produced [¹¹C]methane have now been developed [323, 324]. One procedure involves several rapid recycles of unreacted [¹¹C]methane for iodination [324] and delivers [¹¹C]methyl iodide rapidly in high yield and in high specific activity. This method has become generally preferred [325, 326]. Many of the labeling reactions with [¹¹C]methyl iodide can be performed effectively at room temperature in a procedure, often dubbed the ‘loop method’, in which the labeling agent is passed into narrow bore stainless steel tubing that has been precoated with a small amount of desmethyl precursor [327, 328]. Desmethyl precursors are best prepared de novo rather than by demethylation of the target non-radioactive ligand because even trace impurities of non-demethylated compound may cause carrier dilution in the labeling reaction.

Fig. (19).

Preparation of [¹¹C]methyl iodide and [¹¹C]methyl triflate and utility for radiolabeling various structural motifs.

[¹¹C]Methyl iodide may be converted ‘on-line’ and efficiently into the more reactive ¹¹C-methylating agent, [¹¹C]methyl triflate by passage over heated silver triflate [329]. A much higher-yielding route to [¹¹C] methyl triflate is via [¹¹C]methyl bromide that has been produced from [¹¹C]methane [330]. [¹¹C]Methyl triflate has similar scope to [¹¹C]methyl iodide for labeling at heteroatoms but with lower reaction center selectivity. The use of [¹¹C]methyl triflate in place of [¹¹C]methyl iodide has been noted to improve the practical yields of various PET radioligands [331, 332].

Carbonyl functions appear frequently in small molecule drug-like structures as, for example, in esters and amides. One approach to labeling esters and amides in their carbonyl functions is through the use of ¹¹C-labeled acid chlorides. [¹¹C]Acid chlorides may be prepared in good yield from reactions of cyclotron-produced [¹¹C]carbon dioxide with an organometallic reagent, such as a Grignard reagent, followed by chlorination with a reagent such as phthaloyl dichloride, oxalyl chloride or thionyl chloride [333, 334] (Fig. 20). Volatile [¹¹C]acid chlorides are readily separated from reaction mixtures for subsequent reaction, as in the preparation of [¹¹C]cyclopropanecarbonyl chloride for the radiosynthesis of the opiate receptor radioligand, [¹¹C]diprenorphine [333]. Less volatile [¹¹C]acid chlorides may be produced and used in situ, such as [¹¹C]cyclohexanecarbonyl chloride for the production of the 5-HT_1A receptor radioligand, [carbonyl-¹¹C]WAY-100635 [335, 336]. In the latter case, the labeling agent may be produced conveniently in a narrow polypropylene tube to which it adheres and is made available for reaction with the amine precursor [336]. As already indicated, strenuous precautions are needed to avoid excessive carrier dilution in reactions of [¹¹C]carbon dioxide with the Grignard reagent. For example, a commercial source of reagent may become unacceptably contaminated with carbon dioxide, and it may therefore need to be made freshly in-house with strict precautions against ingress of atmospheric carbon dioxide.

Fig. (20).

Preparation and utility of [¹¹C]acid chlorides.

Carbon monoxide, although generally considered to be quite inert, shows rich chemistry in transition metal-mediated insertion reactions that may generate a variety of compound types, including ketones, acids, esters and amides. [¹¹C]Carbon monoxide may be produced on-line in high yield by passage of [¹¹C]carbon dioxide over heated zinc [337] or molybdenum [337, 338]. A major obstacle to the early application of [¹¹C]carbon monoxide to producing radiotracers was its low solubility in organic solvents. Various approaches have been adopted to surmount this obstacle, including the use of a high-pressure autoclave [339], transient and reversible trapping of the [¹¹C]carbon monoxide in solution [340], and even the generation of the [¹¹C]carbon monoxide from [¹¹C]carbon dioxide in solution [341]. A recent method uses xenon to carry [¹¹C]carbon monoxide efficiently into organic solution at room temperature and low pressure [342]. These advances have opened up the possibility to use [¹¹C]carbon monoxide to label a vast array of carbonyl-containing structural motifs (Fig. 21), often in high yields and without the need for protection-deprotection schemes [343–345]. Moreover, the achieved specific activities are generally very high as there is low risk for contamination of reactions by environmental carbon monoxide. [¹¹C]Carbon monoxide is now increasingly applied to producing PET radioligands (e.g. [¹¹C]FIMX) [346].

Fig. (21).

Synthesis of ¹¹C-labeled ketones, esters, and amides from [¹¹C]carbon monoxide.

Another widely produced labeling agent is [¹¹C] hydrogen cyanide, which is especially useful for preparing aryl [¹¹C]nitriles from transition metal-promoted reactions with aryl halides [347] (Fig. 22), including PET radioligands for CB₁ receptors (e.g. [¹¹C]SD5024) [147] and mGluR5 receptors (e.g. [¹¹C]SP203 [348]). [¹¹C]Hydrogen cyanide is prepared by treating [¹¹C]methane with ammonia gas over heated platinum [147]. Experience has shown that radioligands produced from [¹¹C]hydrogen cyanide generally have lower specific activity than expected from the specific activity of the [¹¹C]methane precursor. A recent study has employed a micro-ion selective electrode to measure the specific activity of [¹¹C]cyanide [349].

Fig. (22).

Preparation of [¹¹C]hydrogen cyanide and application to prepare [¹¹C]aryl nitriles.

Methods are available for preparing a myriad of other secondary labeling agents of as yet less utility and popularity than those mentioned above, including [¹¹C]formaldehyde [350, 351], higher [¹¹C]alkyl halides [352, 353], [¹¹C]nitroalkanes [354], [¹¹C]acetone [355], [¹¹C]phosgene [356–359], and [¹¹C]benzyl halides [360–362]. Readers are directed to the pertinent literature [350–362] for more details on such methods.

14.2. Fluorine-18

Although many methods are known for producing fluorine-18, the ¹⁸O(p,n)¹⁸F reaction on ¹⁸O-enriched water has become almost universally applied, because of its high yield of up to several hundred GBq from modern biomedical cyclotrons [363]. The fluorine-18 is obtained in a single chemical form, namely [¹⁸F]fluoride ion, at very high NCA specific activity. The [¹⁸F]fluoride ion becomes an increasingly reactive nucleophile as water of hydration is removed [296]. In all labeling procedures, care is required to maintain high specific activity by avoiding carrier dilution from materials that may come into contact with the [¹⁸F]fluoride ion, especially for example Teflon tubing [364]. With such precautions, specific activities of radiotracers are typically of the order of 180–370 GBq/μmol at end of synthesis, typically 1 to 2 half-lives after radionuclide production.

Classical aliphatic nucleophilic substitution reactions (notionally S_N1-type) on, for example alkyl bromides or tosylates, may be used to label some radioligands directly, such as [¹⁸F]fallypride [365], [¹⁸F]FBR [106], and [¹⁸F]SP203 [165] (structures shown in Table 4). Aromatic nucleophilic substitutions in homoarenes (notionally S_NAr-type) can usually be achieved on substrates bearing good leaving groups (e.g. Br, NO_2, Me₃N⁺) in ortho or para orientation to a strong electron-withdrawing group (e.g., CO, NO₂, CN). The earlier mentioned radioligand [¹⁸F]XB-1 (Table 4) is an example produced by such a process [159]. For some radioligands, such as [¹⁸F]FPEB (Table 4), this type of procedure can also be adequately effective with a nitrile group in meta orientation [366]. Substituted [¹⁸F]fluoropyridines may also be obtained from [¹⁸F]fluoride ion, by displacement of good leaving groups (Br, NO₂, Me₃N⁺). Activation by an electron-withdrawing group is generally needed for introducing [¹⁸F]fluoride ion into the meta position, but not usually for ortho or para position. These nucleophilic substitution reactions have been extensively reviewed [367, 368].

Over the last two decades, the introduction of hypervalent iodine precursors, such as diaryliodonium salts [369, 370] and aryl ylides [371, 372], has expanded the chemical space that can be labeled with [¹⁸F]fluoride ion to include electron-rich aromatic rings, as recently reviewed [373–375] (Fig. 23). [¹⁸F]Flumazenil [376] and [¹⁸F]FIMX [107] are notable examples of radioligands that have been produced regularly from such hypervalent precursors.

Fig. (23).

Preparation of [¹⁸F]aryl fluorides from hypervalent iodine substrates.

More recently, progress has been made in achieving aryl ring radiofluorination through copper-mediated reactions on substrates, such as diaryliodonium salts [377], aryl boronic acids [378], and aryl boronate esters [379] (Fig. 24). These labeling reactions have wider scope than classical S_NAr reactions, as they are readily applicable to labeling both electron-rich and electron-deficient arenes, irrespective of substitution patterns. The synthetic accessibility of stable arylboron precursors is also attractive.

Fig. (24).

Cu-mediated routes to [¹⁸F]fluoroarenes from [¹⁸F]fluoride ion.

Direct methods for labeling with [¹⁸F]fluoride ion have not always been applicable or preferred. Hence, several methods have been developed for converting [¹⁸F]fluoride ion efficiently into secondary labeling agents, particularly electrophilic agents. Prominent examples are [¹⁸F]dideuterated fluoromethyl bromide and higher functionalized [¹⁸F]fluoroalkanes, such as [¹⁸F]2-fluoroethyl tosylate [380] and bromide [381], and ring radiofluorinated benzaldehydes and benzyl halides [382] (Fig. 25).

Fig. (25).

Examples of conversions of [¹⁸F]fluoride ion into secondary labeling agents.

Many structural motifs that are common in small drug-like molecules have until recently been neglected as possible sites for introduction of a positron-emitter, because of the perceived difficulty of doing so at an NCA level of specific activity. A recent drive in fluorine-18 radiochemistry is now to open up methods for labeling such structural motifs. For example, significant progress has been made in developing methods for labeling alkyl [383] and aryl trifluoromethyl groups [384–387] (Fig. 26).

Fig. (26).

Labeling of alkyl and aryl trifluoromethyl groups with [¹⁸F]fluoride ion.

CONCLUSION

The development of reversibly binding PET radio-ligands for brain imaging poses enormous challenges to chemists and other scientists, and is vulnerable to many factors that may cause failure at any stage. A strong appreciation of the many considerations that are important to such PET radioligand development, as outlined in this article, will hopefully aid chemists to avoid the most obvious pitfalls and to maximize their chances of success. Considerable success is indeed possible as evidenced by the now widespread use of many useful PET radioligands.