Introduction
Several positron emission tomography (PET) ligands exist for the 18-kDa translocator protein (TSPO), which can be upregulated in the mitochondria of activated microglia when the central nervous system becomes inflamed (see [
1] for review). Currently, it is unclear, in humans, what proportion of the observed
in vivo PET signal represents specific TSPO binding and what proportion is merely non-displaceable binding. This is a particular problem with TSPO imaging as (1) the binding of a large majority of TSPO ligands is affected by carriage of a single nucleotide polymorphism (SNP rs6971) and (2) affinity thus varies according to whether participants are homozygotic (high or low affinity binders, HABs/LABs) or heterozygotic (mixed affinity binders, MABs). This results in increased tracer-specific variability across cohorts [
2]. TSPO is known to be expressed ubiquitously throughout the human brain, meaning that there is no suitable reference region (free of specific binding) which would allow the non-displaceable proportion of the signal to be estimated
in vivo (see [
3] for overview). Assuming that the fraction of non-displaceable binding is negligible can dramatically affect the interpretation of results. It is therefore important to investigate the proportion of binding that is specific for each TSPO tracer [
4,
5].
Approaches recently described by Owen et al. [
6] and Guo et al. [
7] can be used to estimate the non-displaceable component of the total volume of distribution (
VND)
in vivo. In short, for the TSPO ligand of interest, the total volume of distribution,
VT, is calculated both before and after blockade with the TSPO ligand XBD173.
VND is derived from the
x-intercept of the graph of
\( {V}_{\mathrm{T}}^{\mathrm{baseline}} \) against
\( {V}_{\mathrm{T}}^{\mathrm{baseline}}-{V}_{\mathrm{T}}^{\mathrm{block}} \) (see ‘
Materials and Methods’). In addition, use of a polymorphism plot (Guo, et al. [
8]), which assumes that MABs express an equal percentage of high and low affinity binding sites [
6,
9], allows
VND to be derived from the
x-intercept of
\( {V}_{\mathrm{T}}^{\mathrm{HAB}} \) against
\( {V}_{\mathrm{T}}^{\mathrm{HAB}}-{V}_{\mathrm{T}}^{\mathrm{MAB}} \) (see ‘
Materials and Methods’).
One recently developed ligand, [
18F]GE-180, has exhibited a higher signal to background ratio than (R)-[
11C]PK11195 in several preclinical models [
10‐
12]. However, in human studies, it has shown unexpectedly low brain penetration [
13‐
15]. Additionally,
in vitro data shows that binding of [
18F]GE-180 to TSPO is sensitive to the presence of the rs6971 SNP [
2,
9,
16]; however, in these
in vivo human studies, the expected genotype dependence of signals was not observed [
13,
14]. This phenomenon may be due to poor extraction of [
18F]GE-180 over the blood−brain barrier (BBB) and/or the action of active efflux pumps such as P-glycoprotein. Given these unexpected results, we wished to clarify the proportion of [
18F]GE-180 uptake detected with PET in the human brain that is non-displaceable. A recent blocking study in healthy control subjects confirmed the presence of specific binding throughout the human brain with [
11C]PBR28 [
6]. Here, we describe a similar blocking study to investigate whether (and what proportion of) [
18F]GE-180 and [
11C]PBR28 PET signal is specific to TSPO binding in people with multiple sclerosis (MS).
Discussion
This study was designed to quantify the non-specific binding (VND) of the TSPO PET tracers [18F]GE-180 and [11C]PBR28 in people with MS. We found that VND accounts for, on average, 55 and 67 % of the total binding of [18F]GE-180 and [11C]PBR28, respectively, indicating that the remaining 45 or 33 % are attributable to specific signal (VS).
[
18F]GE-180 has shown high signal-to-noise ratios in preclinical studies [
10‐
12,
21,
22] but unexpectedly low brain penetration in human healthy controls [
13‐
15]. Recent studies have demonstrated markedly increased uptake of [
18F]GE-180 in people with glioblastoma [
23] and in people with relapsing-remitting MS [
24], but questions remain as to whether this increase in signal represents specific binding or is merely due to non-specific signal in areas of blood−brain barrier breakdown. Another recent study which directly compared [
18F]GE-180 with [
11C]PBR28 in healthy controls who were scanned with both tracers (morning and afternoon) found up to 20 times lower volumes of distribution with the former compared to the latter [
15] as well as difficulties with [
18F]GE-180 quantification. To our knowledge, ours is the first study assessing non-displaceable binding of [
18F]GE-180 and comparing both tracers in disease cohorts. Our results argue that, despite low brain penetration [
13,
14], [
18F]GE-180 does exhibit a specific signal in the MS brain and hence could be useful in conditions with pathologically increased levels of TSPO. We also performed gadolinium contrast-enhancing MRI in the cohort of participants scanned with [
18F]GE-180 and observed no contrast enhancement (in lesion areas or otherwise), suggesting no extensive BBB breakdown. Although this observation does not exclude the possibility of micro-BBB breakdown, which could allow passage of [
18F]GE-180 molecules, but not the larger gadolinium molecules, through the disrupted area, the finding does concur with that of Vomacka et al. [
25]. The authors of this study also comment that (in their previous study [
23]) areas with contrast enhancement in MR did not always correlate with increased [
18F]GE-180 signal, indicating that signal increases in PET are likely to be related to TSPO expression rather than exclusively BBB breakdown. The results also have broader implications on how novel tracers should be validated and compared. While high absolute
VTs are preferable in a tracer, it is crucial to understand what proportion of
VT is driven by
VND. This can be achieved using a blocking study, which, given the lack of an appropriate receptor-free reference region in the brain, we suggest should be undertaken for all TSPO tracers undergoing clinical development.
Although the occupancy plot has been more commonly used in
VND quantification [
4], an alternative approach, relevant for TSPO tracers which are susceptible to the rs6971 SNP, is to create a polymorphism plot, which does not require pharmacological blockade and relies only on the assumption of equal expression of HAB and LAB sites. Our
VND results from both methods (including free and fixed-intercept occupancy plots) were in good agreement for both tracers, giving an average
VND of 0.16 ± 0.05 ml/cm
3 for [
18F]GE-180 and 3.65 ± 0.16 ml/cm
3 for [
11C]PBR28. In this study, we also demonstrate that
VT is consistently greater than
VND for both [
18F]GE-180 and [
11C]PBR28; in other words, no ROIs were devoid of specific TSPO binding. This finding fits with previous observations that reference tissue approaches may not be appropriate in TSPO PET studies [
6].
[
11C]PBR28 has been validated with blocking experiments prior to our study, both in healthy controls [
6,
26] and in a disease cohort [
27]. [
11C]PBR28 is generally accepted as an effective TSPO tracer
in vivo [
28‐
32] although exhibits counterintuitively decreased
VT in subjects with neuroinflammation [
29]. In Owen et al. [
6], the
VND of [
11C]PBR28 was 1.98 ml/cm
3 (~ 50 % of
VT), while Fujita et al. [
26] reported an average
VT of 4.3 ml/cm
3 (in HABs) and a BP
ND of 1.2, giving a very similar
VND of 1.98 ml/cm
3 (~ 45 % of
VT). In our study,
VND for [
11C]PBR28 was 3.65. It is possible that this is a disease-specific difference, given that both the former studies were performed in healthy subjects; however, the sample size in our study was also small and the estimation of
VND may therefore be subject to some biological variability. Nevertheless, the proportion of non-specific binding for [
18F]GE-180 is comparable to or even lower than that of [
11C]PBR28 (
VND ~ 55
vs. ~69 %, respectively), although absolute
VTs are lower [
6,
27]. This result further indicates that [
18F]GE-180 is able to identify specific TSPO signal in the MS brain, in spite of low brain penetration. We also report respective HAB and MAB
VS of 57 and 20 % for [
18F]GE-180 and 37 and 25 % for [
11C]PBR28, although the group numbers (
n = 3 for both [
18F]GE-180 groups and [
11C]PBR28 HABs,
n = 2 for [
11C]PBR28 MABs) are too small to draw firm conclusions.
As has been pointed out previously, the brain penetration of [
18F]GE-180 is very low in humans [
13‐
15]. Our findings also showed low values of
VT and
K1, indicating low extraction of the tracer across the blood−brain barrier (
K1 ~ 0.003
vs. 0.2 ml/cm
3/min for [
11C]PBR28;
i.e., ~ 60× lower). Zanotti-Fregonara and colleagues also noted difficulty in kinetic model fitting of [
18F]GE-180 data using the standard two-tissue compartment model with free blood volume parameter, which we used [
15]. Here, however, we were able to fit the large majority of regions well, with
R2 comparable to those seen with [
11C]PBR28 (
R2 ~ 0.8) and standard errors on
VT estimates < 20 % for both tracers. It is possible that this is due to the larger, less noisy ROIs selected for analysis in our study compared to those used by Zanotti-Fregonara and colleagues. Also to be considered is the fact that our study involved (six) participants with MS, compared to the four healthy controls and single participant with amytrophic lateral sclerosis in the other study, which may have resulted in altered binding kinetics due to disease-specific pathology. Although a methodology considering large ROIs in a disease such as MS, with focal lesion-based pathology, may seem to limit the usefulness of a tracer, there is evidence that there is a global effect on TSPO PET signal in regions such as NAWM and normal appearing grey matter (see [
33] for review), suggesting that [
18F]GE-180 need not be excluded from use based on this fact. Many recent studies using [
11C]PBR28 (and other tracers) have elected to use the 2TCM-1K kinetic model [
34], which incorporates a parameter representing the endothelial fraction of binding of a TSPO tracer, for quantification [
29]. The 2TCM-1K has been used with [
18F]GE-180 data [
13] and has not shown a substantial advantage in terms of parsimony criteria (Akaike Information Criteria, AIC) compared to the 2TCM. Furthermore, [
11C]PBR28 datasets are often still analysed using the 2TCM, primarily for comparison with data from other tracers [
15,
35,
36]. We also found good fits to [
11C]PBR28 data with the 2TCM; thus, here, we elected to use this model to analyse data from both tracers.
The
in vitro HAB/LAB ratio of binding affinity for [
11C]PBR28 has been observed to be approximately 1:50 [
9], while for [
18F]GE-180, this ratio is between 1:5 and 1:15 (personal communication, DRO, WT).
In vivo, [
11C]PBR28-scanned MABs express approximately half the signal compared to HABs [
6]. The slope of the polymorphism plot is equivalent to
\( \frac{{\mathrm{BP}}_{\mathrm{ND}}^{\mathrm{HAB}}-{\mathrm{BP}}_{\mathrm{ND}}^{\mathrm{MAB}}}{{\mathrm{BP}}_{\mathrm{ND}}^{\mathrm{HAB}}} \). For [
18F]GE-180, our equivalent
in vivo HAB/MAB ratio was found to be 5.43 ± 3.27 and the average slope was thus expected to be approximately 0.8, while for [
11C]PBR28, these values were 3.21 ± 1.27 and 0.7. Our results for the slopes of the polymorphism plots were 0.96 and 0.67 respectively and thus fell within one standard deviation of the predicted values. Previous studies using [
18F]GE-180 have been unable to detect consistent differences in tracer signal between HABs and MABs [
13,
14,
24], indicating that brain penetration of the tracer is low, except where BBB breakdown may be present [
24]. Contrary to these results, in our study, HABs exhibited approximately twice the total volume of distribution of MABs in selected ROIs at baseline. It is likely that the difference in our study is driven primarily by one HAB with particularly high signal (participant D) and one MAB with particularly low signal (participant E). Clearly there is considerable variability in population
VT and with the small sample size in our study (
n = 6), we are unable to validate our findings with statistical tests. In addition, the difference in scan timing between the first and second (post-XBD173) scan varied for each tracer (half a day for [
11C]PBR28 and 1 week for [
18F]GE-180), which itself may have introduced some variability in PET signal. Thus, we suggest continuing binding status stratification of participants in future studies, where larger data pools may enable more reliable HAB/MAB binding ratio estimates.
Several caveats exist in this study. Firstly, the occupancy of XBD173 in the two cohorts (slopes in the occupancy plots) varied between 24 and 95 % (fixed-intercept plot) for [
18F]GE-180 and 16 and 220 % for [
11C]PBR28. Clearly, the occupancy of XBD173 cannot exceed 100 %, and indeed, the dose administered was pre-calculated to give an expected approximate 75 % occupancy, allowing for participant weights and considering the dose-occupancy relationship described by Owen et al. [
6] for [
11C]PBR28. Of course, it should also be noted that these values were obtained from the fixed-intercept occupancy plot; for the free intercept plot, occupancies were within one standard deviation of the expected value. This point again highlights the large variability across cohorts and the fact that averaging results across participants, even within the same disease population, may not be a suitable approach. In addition, one participant scanned with [
11C]PBR28 exhibited no substantial occupancy and was excluded from further analysis. This large inter-individual variability in occupancy was also seen in previous XBD173 blocking studies [
6,
27], reflected in the variability in total signal reduction (
VT) between pre and post-block scans. Whether driven by biological variability or experimental noise, these results provide evidence that
VND cannot necessarily be assumed to be the same across regions and a population. If there is indeed a biological spectrum of
VND between individuals, this raises the interesting possibility that blocking scans should be included for all participants in all TSPO PET studies to optimally quantify
VS. This would have broad repercussions on TSPO PET study design, including cost, radiation exposure and participant discomfort. All neuro-PET tracers are, of course, better able to penetrate brain tissue when the BBB is disrupted. In the case of [
18F]GE-180, which exhibits low penetration of the healthy BBB, this is particularly relevant. Although our cohorts were selected due to their clinically low or ‘inactive’ MRI, and although we investigated only lesion-free ROIs, we only performed contrast enhanced MRI to estimate BBB integrity in the [
18F]GE-180 cohort, and no measure of micro-BBB disruption was performed. Lastly, although this study was performed in a cohort of people with MS, we have not investigated how
VT differs in MS lesions or how it correlates with clinical outcomes. This study was not powered to address these questions, since participants were not burdened with large lesion loads. Instead, these questions will be explored in follow-up studies.