Sage Journals: Discover world-class research

Abstract

The 18 kDa translocator protein (TSPO) is a marker of microglia activation and the main target of positron emission tomography (PET) ligands for neuroinflammation. Previous works showed that accounting for TSPO endothelial binding improves PET quantification for [¹¹C]PBR28, [¹⁸F]DPA714 and [¹¹C]-R-PK11195. It is still unclear, however, whether the vascular signal is tracer-dependent. This work aims to explore the relationship between the TSPO vascular and tissue components for PET tracers with varying affinity, also assessing the impact of affinity towards the differentiability amongst kinetics and the ensuing ligand amenability to cluster analysis for the extraction of a reference region. First, we applied the compartmental model accounting for vascular binding to [¹¹C]-R-PK11195 data from six healthy subjects. Then, we compared the [¹¹C]-R-PK11195 vascular binding estimates with previously published values for [¹⁸F]DPA714 and [¹¹C]PBR28. Finally, we determined the suitability for reference region extraction by calculating the angle between grey and white matter kinetics. Our results showed that endothelial binding is common to all TSPO tracers and proportional to their affinity. By consequence, grey and white matter kinetics were most similar for the radioligand with the highest affinity (i.e. [¹¹C]PBR28), hence poorly suited for the extraction of a reference region using supervised clustering.

Keywords

Translocator protein kinetic modelling neuroinflammation microglia endothelium clustering reference region

Introduction

The translocator protein 18 kDa (TSPO) is a five-transmembrane domain protein located in the outer mitochondrial membrane.¹ During the inflammatory response in the brain, TSPO is upregulated in microglial cells and macrophages and, to a lesser extent, in reactive astrocytes.^2–4 Therefore, TSPO is a target of interest for positron emission tomography (PET) studies aimed at imaging neuroinflammation in vivo.^5,6

The quantification of TSPO density with PET is a challenging task with any radioligand.^7,8 Firstly, a single nucleotide polymorphism (rs6971) in the TSPO gene affects the binding affinity of second-generation TSPO tracers.⁹ Secondly, all TSPO radiotracers can bind to a number of sites in blood (e.g. platelets and monocytes, but also plasma proteins), and this binding can be altered in pathological conditions affecting the peripheral immune system.^10–15 Given that the free fraction in plasma of TSPO radioligands is generally very low (between 1% and 5%) and difficult to measure,^8,16 quantification with plasmatic input function is challenging. By consequence, reference tissue input models are generally preferable over plasma input models. However, the widespread distribution of TSPO in the normal brain and the unknown pattern of TSPO density in pathological conditions prevent the use of anatomically defined regions of reference. The cerebellum, often used as a reference or pseudo-reference for many PET tracers, including [¹¹C]PBR28,¹⁷ might not be an ideal reference region for TSPO PET due to its complex tissue composition¹⁸ and its unpredictable susceptibility to inflammation.¹⁹

One approach for non-invasive quantification of [¹¹C]-R-PK11195 scans consists of modelling the kinetics in each voxel as a linear combination of kinetics of predefined classes, and in defining the reference as the average contribution of the normal grey matter in each voxel. This approach, which is a form of supervised clustering, is known as super-PK^19,20 and was recently developed and validated also for [¹⁸F]DPA714 data.²¹

The last challenge to quantification is related to the heterogeneous TSPO distribution in the healthy brain and its impact on kinetic modelling.^18,19 Although the background signal of TSPO radiotracers seems homogeneous on PET images, at the microscopic level, TSPO is overexpressed in the endothelium and the tunica media of brain vessels.¹⁹ Saturation binding studies and immunohistochemical studies have repetitively confirmed it in different conditions.^22–25 This TSPO fraction, which henceforth will be referred to as vascular or endothelial fraction, is reduced in aged brains and during neurodegenerative conditions due to vascular fibrosis.²⁶ A first study with [¹¹C]-R-PK11195 showed that incorporating the vascular TSPO component into a reference-based model (simplified reference tissue model with vascular component, SRTM-V_b) improved the differentiation of Alzheimer’s disease patients from controls.²⁶ Compared to reference-based models without endothelial correction, SRTM-V_b proved to be superior for the quantification of brain TSPO density with [¹¹C]-R-PK11195 in showing differences in binding between subject groups.²⁰ We recently proposed a two-tissue compartmental model with vascular binding (2TCM-1K) for the quantification with a plasma input function of the high-affinity [¹¹C]PBR28 that explicitly accounts for the endothelial TSPO binding.²⁷ This model improved the quantification of brain [¹¹C]PBR28 data, both in healthy populations and in subjects affected by schizophrenia;^27,28 similar improvements were gained by its recent application to [18F]DPA714 data.²⁹

For the present work, we postulated that the contribution of vascular TSPO to the total TSPO binding would be proportional to the affinity of the tracer, i.e. high-affinity second-generation compounds would be more affected by vascular TSPO binding than tracers with lower affinity, such as [¹¹C]-R-PK11195. In particular, when using high-affinity compounds, higher vascular and parenchymal signal would actually lower tissue contrast making the extraction of a reference region by clustering arduous.

Hence this study sought to (a) validate the vascular-trapping model for [¹¹C]-R-PK11195; (b) determine the consistency of the results with those obtained by reference model SRTM-V_b; (c) calculate the relationship between the estimated endothelial binding rate values of [¹¹C]-R-PK11195, [¹⁸F]DPA714²⁹ and [¹¹C]PBR28²⁸ and their affinity and (d) calculate the ratio between parenchymal grey and white matter signals as an index to determine what tracers are amenable to quantification with a pseudo-reference region obtained by cluster analysis.

Materials and methods

Vascular binding model for [¹¹C]-R-PK11195

Dataset

The dataset consisted of six healthy subjects from previously reported studies.^19,26 The studies were approved by the Hammersmith Hospitals Trust Ethics Committee and were conducted in accordance with the Declaration of Helsinki. Permission to administer radioisotopes was granted by the Administration of Radioactive Substances Advisory Committee of the Department of Health, U.K. Informed written consent was obtained from all subjects.

PET imaging

All [¹¹C]-R-PK11195 studies were performed on an ECAT EXACT 3D (CTI/Siemens) PET camera, with a 23.4-cm axial field of view, 95 transaxial planes and a spatial resolution of 4.8-mm (transaxial) and 5.6-mm (axial) full width at half maximum (for details, see Spinks et al.³⁰). A transmission scan was acquired before every emission using a single rotating photon point source of 150 MBq of ¹³⁷Cs for subsequent attenuation correction and scatter correction. Thirty seconds after the start of the emission scan, the subjects received a bolus injection of [¹¹C]-R-PK11195 (296 MBq) followed by a 60-minute PET emission scan. PET data were in list mode and rebinned as 18 time frames (durations: 30-s background frame, 1 × 15 s, 1 × 5 s, 1 × 10 s, 1 × 30 s, 4 × 60 s, 7 × 300 s, 2 × 600 s). Data were reconstructed using the filtered backprojection algorithm.³¹ Reconstructed voxel sizes were 2.096 mm × 2.096 mm × 2.43 mm. Radiopharmaceutical preparation of [¹¹C]-R-PK11195 was done as described in Tomasi et al.²⁶

In addition, volumetric T1-weighted MR images were obtained on a 1.0-T Picker HPQ scanner (Picker International) at the Robert Steiner MR Unit, Hammersmith Hospital, London. These structural MRI images were used along with PET imaging for region segmentation.

Arterial blood sampling and analysis of [¹¹C]-R-PK11195 plasma metabolism

Analysis of blood data is described in detail in the original reference.¹⁹ Briefly, arterial blood was sampled during PET acquisition via the radial artery using a combined automatic–manual approach. Arterial whole-blood activity was monitored continuously for the first 15 min of the scan at a flow rate of 5 mL/min.³² Eight discrete arterial blood samples were manually withdrawn at 5, 10, 15, 20, 30, 40, 50 and 60 min into heparinized syringes. The activity concentrations of the whole blood and plasma were used to determine the plasma-over-blood activity ratio, which was fitted with a straight line.¹⁹ Samples at 5, 10 and 15 min were used to calibrate the two sampling modalities. Five plasma samples per scan (at 5, 10, 20, 40 and 60 min) were analysed to calculate the parent fraction in plasma (PPf).

Processing of blood samples was performed consistently with previous studies.¹⁹ PPf were fitted with an exponential model, while whole blood time activity curves (TACs) were fitted using a variation of Feng’s model that consists in a straight line until the arterial input function peak, followed by a tri-exponential decay.³³ For each scan, the time difference between tracer arrival to the brain and to the arterial sampling site was compensated for by shifting the blood curves by 0–20 s (both parent and whole blood TACs), fitting the whole brain TAC (using linear spectral analysis to avoid dependency on a particular compartmental model³⁴), and selecting the delay that produced the smallest weighted residual sum of squares.

Image analysis

Structural MRI images were used for grey and white matter segmentation and region of interest (ROI) definition. The Hammersmith maximum probability atlas³⁵ of 73 ROIs was co-registered on each subject’s MRI scan and PET image using FSL (http://www.fsl.fmrib.ox.ac.uk/fsl) functions. ROIs included whole brain, occipital lobe, temporal lobe, frontal lobe, parietal lobe, insula, cingulate cortex, amygdala, hippocampus, thalamus, caudate, putamen and cerebellum.

All PET images were corrected for head movement by denoising the non-attenuation-corrected dynamic images with a Battle-Lemarie wavelet filter (level 2, order 64). Non-attenuation-corrected images were used, because the scalp signal is stronger and that improves re-alignment compared to attenuation-corrected images. All frames were realigned to a single “reference” space corresponding to the PET frame with the highest activity. The transformation parameters were then applied to the corresponding attenuation-corrected PET frames to create a movement-corrected dynamic image for analysis. Regional TACs were obtained by sampling the image with the co-registered atlas.

[¹¹C]-R-PK11195 kinetic analysis

Quantification of [¹¹C]-R-PK11195 tissue distribution was performed at the region level by comparing different compartmental models: two models accounting for tissue binding only (two-tissue and one-tissue compartmental models in their standard definition – 2TCM and 1TCM, respectively) and two models that include the vascular TSPO binding (2TCM-1K and 1TCM-1K, respectively). A summary of the model configurations and kinetic parameters is reported in online supplementary Figure 1. Also note that the models share most of the micro-parameters with the only exception of K_b [1/min], which is explicitly used in 2TCM-1K and 1TCM-1K to account for the vascular component.²⁷

Figure 1.

[¹¹C]PK11195 brain PET data fit and model compartmentalization. Example of model fits to whole brain activity with standard 1- and 2-tissue modelling (a, b) and including vascular modelling with 1TCM-1K (c) and 2TCM-1K (d) in a representative healthy subject. In each panel, black circles represent the measured activity while the solid black lines represent the data model description. Single compartmental kinetics is described by blue, green and red lines, which represent non-displaceable, specific and endothelial binding, respectively. Purple solid lines represent total tissue binding (i.e. non-displaceable plus specific binding).

We also analysed the data using the modified version of the simplified reference tissue model (SRTM-V_b) proposed in Tomasi et al.²⁶ This modified version accounts for tracer vascular activity both in reference and target tissues, using the whole vasculature TAC extracted from the image to correct for the endothelial TSPO binding. The reference region was derived with the super-PK algorithm.¹⁹

All compartmental models were analysed using the weighted nonlinear least square estimator (WNLLS), as implemented in Matlab (The Mathworks Inc., Natick, MA). Initial microparameter values were set based on the literature.^19,27 The weights for the individual data points were defined as the inverse of the variance of the PET measurement error. This was assumed to be additive, uncorrelated, normally distributed with a zero mean and variance equal to the decay-corrected activity divided by the length of the relative scan interval, multiplied by a scale factor estimated a posteriori.³⁶ We calculated the total distribution volume (V_T, mL/cm³) for 1TCM, 1TCM-1K, 2TCM and 2TCM-1K. The volume of distribution is the ratio of the concentration of radioligand in a region of tissue to that in plasma.³⁷ Regional values of V_T estimated with low precision (CV > 50%) were considered outliers and excluded from further analysis.²⁷ For all models, the blood volume fraction (V_b) was estimated together with the other model microparameters.

Model selection was performed using the Akaike information criterion (AIC), which balances model accuracy versus model complexity,³⁸ and by calculating the percentage of outliers. The model descriptions of the tissue data were also analysed in terms of weighted residuals, which had to be randomly distributed and uncorrelated, following the assumptions on the model error. We applied the Wald–Wolfowitz runs test for the randomness of the residuals and a one-sample Kolmogorov–Smirnov test to assess if residuals came from a standard normal distribution.

All plasma input models were compared with the simplified reference tissue model SRTM-V_b on the basis of the estimates of the distribution volume ratio ( $DVR$ , unitless), calculated for each region as the ratio of the regional V_T and that of the reference grey matter (defined with the super-PK). Since the non-displaceable binding potential (i.e. the ratio at equilibrium of specifically bound radioligand to that of non-displaceable radioligand in tissue) can be expressed as $DVR - 1$ , DVR reflects the tracer specific binding to TSPO receptors.³⁷ $DVR$ estimates obtained with input function models and SRTM-V_b were compared in terms of percentage relative difference with respect to the arithmetic mean and Pearson’s linear correlation coefficient.

Relationship between [¹¹C]-R-PK11195, [¹⁸F]DPA714 and [¹¹C]PBR28 vascular component and tracer affinity

We speculated that the magnitude of the vascular component would be proportional to tracer affinity, i.e. the higher the affinity, the higher the values of K_b. In order to assess the relationship between the vascular component and tracer affinity, we compared the estimates of K_b (i.e. the endothelial tracer binding) in three different tracers with increasing affinity for the TSPO: [¹¹C]-R-PK11195, [¹⁸F]DPA714 and [¹¹C]PBR28.³⁹ The affinity of [¹⁸F]DPA714 and [¹¹C]PBR28 is ∼1.5 fold and ∼5 to 6 fold that of [¹¹C]-R-PK11195,³⁹ respectively, with the following reported values: K_i = 4.5–22.3 nM for [¹¹C]-R-PK11195, K_i = 7.0 ± 0.4 nM for [¹⁸F]DPA714⁴⁰ and K_i = 2.5–4.0 nM for [¹¹C]PBR28.⁴¹

The [¹¹C]-R-PK11195 K_b estimates of the six healthy subjects of the present study were compared to the [¹¹C]PBR28 estimates reported by Bloomfield et al. in 18 healthy subjects²⁸ and to the [¹⁸F]DPA714 estimates obtained with the 2TCM-1K in five healthy subjects, part of an ongoing study for the investigation of neuroinflammation in multiple sclerosis. The [¹¹C]PBR28 results were reproduced from the original publication for the sake of this analysis, whereas [¹⁸F]DPA714 data were reanalysed using the Hammersmith maximum probability atlas to derive a set of regions consistent with [¹¹C]PBR28 and [¹¹C]PK11195 datasets.

The experimental protocol for the latter study is described in Lavisse et al.⁴² We limited the analysis to high-affinity binder (HAB) subjects to avoid the inclusion of additional variables in the affinity comparison. The K_b values were compared in the regions that were common to the three datasets, i.e. whole brain, hippocampus, thalamus, caudate, putamen, frontal, parietal, occipital and temporal lobes. The cerebellum was excluded from the comparison, because of its complex cellular heterogeneity and its high content of white matter, which has a higher vessel density per volume unit.¹⁸ Differences in K_b values among tracers were statistically analysed with a repeated-measure ANOVA using SPSS version 21 (http://www.spss.com).

The role of TSPO tracer affinity in using supervised-clustered reference region approaches

The super-PK method relies on three main steps: (1) an input normalization procedure to scale each volume of the dynamic PET sequence, (2) the derivation of a set of predefined kinetic classes and (3) a regression analysis to calculate the contribution of each class to the single voxel kinetic. The fundamental step is the definition of the vectors of the kinetic classes that, albeit non-orthogonal, need to be as independent as possible.¹⁹ These classes are defined anatomically from the segmented MR and PET images and include grey and white matter, blood, skull, muscle and a high binding region. To the authors’ knowledge, there have not been successful implementations of supervised clustering approaches to second-generation TSPO tracers, other than a recent application to [¹⁸F]DPA714.²¹

In order to assess the relationship between the affinity and a successful implementation of supervised clustering, we measured the angle between the kinetic vectors of the grey matter and white matter classes derived for [¹¹C]-R-PK11195 data according to Turkheimer et al.¹⁹ We also included the same trigonometry measurements for [¹¹C]PBR28 and for [¹⁸F]DPA714 data (18 and 5 healthy HAB subjects, respectively, see literature^21,28). The angle between classes was calculated as $ϑ = acos (\bar{GM} \cdot \bar{WM})$ , where the operator $acos$ indicate the inverse of the cosine function, the symbol ċ indicates the dot product, and $\bar{GM}$ and $\bar{WM}$ represent the population-averaged grey and white matter TACs normalized in z-score, respectively. Note that for angle values close to 0°, the kinetics of the classes are the same, whereas the orthogonality is reached for vectors angled at 90° (no similarity between grey and white matter).

We speculated that there would be a negative association between the suitability to reference tissue extraction by clustering and the affinity for TSPO of these ligands. An increase in contrast between parenchymal tissues (grey and white matter) would be expected with increased affinity, but it would be counteracted by the parallel increase of vascular binding. This would ultimately reduce tissue contrast, thus preventing the implementation of a supervised clustering approach and the extraction of a pseudo-reference region.

Results

[¹¹C]-R-PK11195 PET kinetic analysis

Comparison of model performance

The vascular models (either 2TCM-1K or 1TCM-1K) provided a more parsimonious data description in the majority (77%) of the ROIs (Figure 1). Note that 2TCM consistently described the last part of the tissue curve by fitting the specific compartment as an almost irreversible kinetic. The very slow specific component was due to small values of k₄, close to zero (median k₄ = 0.004 min⁻¹). Weighted residuals obtained with all models (except with 1TCM) were consistent with the assumptions that the measurement error was random and uncorrelated. These findings are consistent with the performance of vascular models when applied to other TSPO PET tracers including [¹¹C]PBR28 in healthy and pathological populations^27,28 and [¹⁸F]DPA714 data.²⁹

In terms of parsimony criteria, the 1TCM-1K AIC was lowest in 63% of the brain regions where all models successfully converged to a physiological solution. Based on this result, we selected 1TCM-1K as the optimal model to describe [¹¹C]-R-PK11195 brain data using an arterial blood input function (online supplementary Table 1).

Table 1.

Mean and variability of DVR regional estimates between subjects.

ROIs	1TCM	2TCM	1TCM-1K	2TCM-1K	SRTM-V_b
Whole brain	1.43±0.12	1.07±0.10	0.89±0.01	0.87±0.04	1.25±0.05
Hippocampus	1.49±0.17	0.56±0.55	1.08±0.23	1.05±0.22	1.27±0.07
Amygdala	1.52±0.22	0.52±0.39	0.76±0.00	0.94±0.18	1.23±0.11
Cerebellum	1.40±0.12	1.18±0.08	0.99±0.06	0.94±0.07	1.24±0.06
Insula	1.60±0.21	0.64±0.53	1.18±0.08	1.12±0.10	1.36±0.11
Thalamus	1.72±0.24	1.03±0.00	1.40±0.17	1.46±0.18	1.47±0.13
Caudate	1.30±0.36	0.09±0.02	0.80±0.10	0.82±0.08	1.14±0.25
Putamen	1.64±0.21	0.52±0.65	1.24±0.11	1.20±0.11	1.45±0.10
Frontal lobe	1.46±0.15	0.85±0.44	0.89±0.03	0.85±0.06	1.30±0.09
Parietal lobe	1.42±0.11	1.14±0.14	0.80±0.02	0.78±0.07	1.27±0.05
Occipital lobe	1.40±0.13	0.84±0.45	0.86±0.05	0.84±0.05	1.26±0.05
Temporal lobe	1.41±0.14	0.98±0.14	0.95±0.05	0.92±0.06	1.24±0.06

Distribution volume ratio ( $DVR$ , unitless) of [¹¹C]-R-PK11195 brain data obtained with 1TCM, 2TCM, 1TCM-1K, 2TCM-1K and SRTM-V_b are reported for several ROIs as mean±SD among six subjects.

As expected, the more complex the model, the higher the fraction of outliers (58% for 2TCM-1K, 48% for 2TCM and no outliers for 1TCM-1K and 1TCM). Notably, for 2TCM, the outliers were always due to very small (close to zero) estimates of k₄ (median k₄ = 0.004 min⁻¹), which suggested almost irreversible kinetics. In the regions where 2TCM was estimated reliably, the median value of k₄ was 0.027 min⁻¹.

Once corrected for outliers, the precision of individual V_T estimates were higher for more complex models, as expected, but in general were consistent among models: for 2TCM1-1K $CV V_{T}$ = 19% ± 11%; for 2TCM $CV V_{T}$ = 18% ± 12%; for 1TCM-1K $CV V_{T}$ =12% ± 8% and for 1TCM $CV V_{T}$ = 5% ± 3%.

Kinetic estimates

The inclusion of a vascular binding compartment resulted in a reduction of the parenchymal V_T by 68% ± 9% (comparing 1TCM-1K and 2TCM estimates). The regional variability of V_T (SD of regional values between subjects normalized by the mean) was comparable with the two models (22% for 1TCM-1K and 19% for 2TCM, respectively). Similarly to the results obtained with [¹¹C]PBR28,²⁷ the V_T values obtained with 2TCM and 1TCM-1K were not only significantly different (paired t-test, p-value<0.001), but also poorly correlated, with an $R < 0.01$ . On the contrary, the values of V_T obtained with the two models including the vascular binding (1TCM-1K and 2TCM-1K) were correlated ( $R = 0.97$ , p-value<0.0001), with a mean relative difference of −3% ± 3%.

When we compared 2TCM and 1TCM-1K $DVR$ estimates with the reference-based model values, SRTM-V_b $DVR$ estimates had the best agreement with 1TCM-1K $DVR$ (mean relative difference, 17% ± −11%, compared to 37% ± 39% differences with 2TCM $DVR$ ). The results of $DVR$ quantification are also reported in Table 1 and in online supplementary Figure 2.

Figure 2.

Mean and variability of regional K_b estimates. TPSO vascular binding estimates (K_b) obtained from tracers with different TSPO affinity. K_b estimates (mean±SD) are shown for [¹¹C]-R-PK11195 (blue bars, n = 6), [¹⁸F]DPA714 (orange bars, n = 5) and [¹¹C]PBR28 data (grey bars, n = 18).

SRTM-V_b $DVR$ estimates were well correlated with 1TCM-1K $DVR$ ( $R = 0.48$ , p-value<0.001), whereas there was no correlation with 2TCM $DVR$ values ( $R = 0.08$ , p-value = 0.52). These results confirmed the equivalence of the vascular correction approaches for [¹¹C]-R-PK11195 kinetics, either by explicitly modelling the endothelial component as a compartment (plasma input modelling) or by using the vascular signal extracted from the image (reference input modelling).

The endothelial fraction was consistent across the brain and precisely estimated (average precision: 16% ± 9%). The between-subject mean value of K_b in the whole brain was 0.14 ± 0.02 L/min (Figure 2 and online supplementary Table 2).

Comparison of the vascular component for tracers with different affinity

The endothelial binding rate estimates for [¹¹C]-R-PK11195, [¹⁸F]DPA714 and [¹¹C]PBR28 are reported in Figure 2 and online supplementary Table 2. The vascular binding K_bestimates were statistically different among groups (repeated measure ANOVA, p-value = 0.02), and the differences held for any region considered (p-value<0.05, Bonferroni corrected), but for the putamen, frontal and parietal lobes. The vascular component was proportional to tracer affinity: the higher the tracer affinity (in our case, [¹¹C]PBR28> [¹⁸F]DPA714>[¹¹C]-R-PK11195), the higher the K_b estimates. The between-subject mean value of K_b in the whole brain was 0.16 ± 0.07 L/min for [¹⁸F]DPA714 and 0.24 ± 0.08 L/min for [¹¹C]PBR28.

Therefore, the affinity of the tracers was positively related to the magnitude of the vascular component.

Similarity of tissue classes kinetics

The normalized grey and white matter kinetics are reported in Figure 3. [¹¹C]PBR28 displayed the most similar curves (angle between grey and white matter kinetics equal only to 5°), whereas the similarity of grey and white matter kinetics was the lowest for [¹¹C]-R-PK11195 (angle of 21°). [¹⁸F]DPA714 data presented an intermediate value (angle of 13°). In summary, as hypothesized, the angles amongst kinetic classes were negatively associated to the tracer affinities.

Figure 3.

Comparison of grey and white matter tissue kinetics. Analysis of the orthogonality of grey and white matter tissue kinetics for [¹¹C]-R-PK11195 (a), [¹⁸F]DPA714 (b) and [¹¹C]PBR28 data (c). The kinetics are reported in z-score with respect to the whole brain activity and normalized by the maximum as in supervised reference tissue clustering.¹⁹ z-Score normalization was calculated by subtracting from each frame its mean and dividing it by its standard deviation. White matter kinetics are displayed in the figure as the inversed curve (i.e. multiplied by −1), in order to highlight the similarity between the grey and white matter time courses.

Discussion

TSPO vascular binding correction: Reference region or blood input function?

In this study, we extended the plasma input kinetic model including vascular binding to [¹¹C]-R-PK11195 data and we showed that accounting for the endothelial TSPO component provides a statistically superior description of the tracer uptake, compared to the standard tissue kinetic modelling without correction. This suggests that the endothelial binding model is applicable to TSPO tracers in general, regardless of their affinity for TSPO.

In particular, the 1TCM-1K model provided a more parsimonious data description and better agreement with the reference SRTM-V_b estimates than the compartmental models not including vascular binding. Notably, a similar data description has already been reported in the literature for [¹¹C]-R-PK11195 using spectral analysis (see Figure 5 in Turkheimer et al.¹⁹). This finding is also consistent with previous studies showing that [¹¹C]-R-PK11195 binding to endothelial TSPO must be corrected for when considering reference region-based modelling.^20,26

Importantly, in the input function-based model, there is an explicit compartmentalization of the vascular component (Figure 4(a)), which is corrected for at the region level. By contrast, the modified SRTM-V_b performs a global correction of the overall tracer activity in the vasculature (which includes the radiotracer both in blood and bound to the blood-brain barrier), obtained by deriving the whole vasculature TAC from the image ( $C_{B}^{IDIF}$ ) from the 10 voxels with the highest activity in the first five frames (Figure 4(b)). The two vascular corrections are equivalent, as $C_{B}^{IDIF}$ can be thought as the sum of the tracer kinetic measured invasively in the whole blood (C_b) and the vascular compartment $C_{vasc}$ of the 1TCM-1K, modulated by the vascular binding rate K_b and the vascular volume fraction ( $C_{B + vasc}$ , Figure 4(c)). The calculations to derive $C_{B + vasc}$ are reported in online supplementary material. Thus, the two approaches share the same rationale, and the blood volume parameter V_b has the same meaning, representing the impact of the vascular fraction on the tissue kinetics (Figure 4(d)). The V_b values of SRTM-V_b and 1TCM-1K showed high correlation ( $R = 0.8$ , p-value < 0.01) and excellent agreement (mean relative difference = −2% ± 19%). Ideally, there should be perfect agreement between the two methods; however, SRTM-V_b performs a global correction, whereas the 1TCM-1K corrects for the vascular binding region-wise, thus modulating the vascular contribution locally through the parameter K_b. As expected, the correlation between SRTM-V_b and 1TCM-1K $DVR$ increases when the 1TCM-1K is solved by fixing K_b to an average value, therefore employing a global correction (data not shown).

Figure 4.

Correcting for vascular TSPO using either blood input functions or reference-based models. The 1TCM-1K and SRTM-V_b approaches for vascular binding correction are illustrated in (a) and (b), respectively. In the input function model 1TCM-1K approach, the tracer activity in the vasculature ( $C_{vasc}$ ) is explicitly compartmentalized and modulated region-wise by the vascular binding rate K_b. By contrast, the SRTM-V_b approach performs a global correction of the overall tracer activity in the vasculature by extracting information on the tracer activity in both the whole blood and endothelial binding from an image-derived whole vasculature TAC ( $C_{B}^{IDIF}$ ). Despite the differences, the two corrections provide an equivalent description of the vasculature component, because $C_{B}^{IDIF}$ corresponds to the sum of the tracer activity in the whole blood (C_b) and in the vascular compartment $C_{vasc}$ (c). With both approaches, V_b represents the weight of the contribution of the vascular component to the total tracer uptake in a given volume of observation (d). Although not perfectly identical (mean relative difference = −2% ± 19%), the V_b estimates are in high agreement (R = 0.8). The results refer to 1TCM-1K and SRTM-V_b, applied to the [¹¹C]PK11195 PET dataset.

Vascular binding in TSPO tracers

In this study, we extended to [¹¹C]-R-PK11195 our previous results obtained by modelling the vascular component for [¹¹C]PBR28 and [¹⁸F]DPA714. We have previously demonstrated that the inclusion in the kinetic model of a vascular compartment to account for the tracer binding to TSPO in vessels (i.e. 2TCM-1K) improves the quantification of brain [¹¹C]PBR28 data, both in healthy populations and in subjects affected by schizophrenia.^27,28 The improvement was as well observed in [¹⁸F]DPA714 data on healthy volunteers.²⁹ Moreover, the use of 2TCM-1K for brain [¹¹C]PBR28 PET images has been recently validated with a displacement study with TSPO ligand XBD173.¹⁸ This study showed that only by including the vascular compartment it was possible to apportion the signal change due to the competing agonist to the correct specific compartments.¹⁸

It should be underscored that the [¹¹C]PK11195 dataset presented in the current study includes only healthy subjects, who in general have no high levels of activated microglia and therefore high TSPO binding. However, as previously shown with [¹⁸F]DPA714 and [¹¹C]PBR28 data, the inclusion of a vascular component in the kinetic model improves the description of brain [¹¹C]PK11195 data also in healthy volunteers. Notably, the 2TCM-1K model, which is preferable for [¹⁸F]DPA714 and [¹¹C]PBR28 data, returned a high percentage of outliers when applied to [¹¹C]PK11195 kinetics. As a consequence, the model that was eventually chosen was the simpler 1TCM-1K. It is possible that the low specific binding in the healthy subjects prevented the identification of a more complex model such as 2TCM-1K. The model 2TCM-1K might be applicable to [¹¹C]PK11195 PET data from subjects with high levels of neuroinflammation.

A key aspect related to the inclusion of a vascular binding in the kinetic model for TSPO tracer is the apparent inconsistency between the reversible binding of the tracer to the parenchyma and the irreversible binding to the vasculature. As we explained previously, this is due to the different reactive volumes, i.e. while the free plasma concentration of the tracer exchanges with endothelial TSPO, only the free fraction in tissue that reaches the mitochondria exchanges with TSPO in microglia and astrocytes.²⁷ This difference is not properly modelled by the common compartment of the 2TCM model, where the free tracer concentration is the average of a rapidly decreasing concentration from the blood-brain barrier to the intracellular space and mitochondria.⁴³ Moreover, modelling the endothelial TSPO as a reversible binding in the framework of a 2TCM does not guarantee the identifiability of the parameter (see supplementary materials in Veronese et al.¹⁸). The inclusion of a vascular component in parallel to the standard 2TCM was originally supported by the known presence of TSPO in brain vessels confirmed by several studies,^22,25 but other scenarios could locate some apparently irreversible component in tissue where slow transport processes and gradients may prevent effective removal. These alternatives should be appropriately tested, since it is unlikely to be determined simply by kinetic modelling of the brain tissue data.

By comparing the vascular binding estimates for TSPO tracers with different affinities, we confirmed our hypothesis that high-affinity second-generation compounds are more affected by vascular TSPO binding than the tracers with lower affinity such as [¹¹C]-R-PK11195. The heterogeneous cellular distribution of TSPO in brain not only makes the tissue quantification particularly challenging but also complicates the definition of a proper reference region, given the ubiquity of TSPO in normal brain and the fact that the pattern of TSPO up-regulation is generally unknown.

Supervised clustering¹⁹ has been the method of choice for a number of studies to automatically define the reference region for [¹¹C]-R-PK11195 data. However, this approach has never been replicated for any second-generation TSPO tracers, other than for [¹⁸F]DPA714.²¹ In this work, we hypothesized that the tracer affinity, by increasing both parenchymal and vascular binding, would modulate the contrast between tissue kinetics, and thus affect the use of cluster analysis to determine a pseudo-reference region. By comparing the contrast between grey matter and white matter kinetics, we found that [¹¹C]PBR28 displayed the lowest tissue contrast (which suggests that high-affinity tracers are poorly suited to supervised reference tissue analysis), whereas middle-affinity tracers such as [¹⁸F]DPA714 would be amenable to supervised definition of a reference region (as confirmed by the successful implementation of super-DPA²¹).

Another limitation of the study is that the three tracers were acquired in different groups of subjects and with different experimental designs. As a consequence, our contrast analysis was performed at population level only. We tested the sensitivity of the tissue kinetics similarity measures by running some tests in an independent brain PET dataset, consisting of eight healthy subjects who were imaged twice on a high-resolution research tomograph (HRRT) scanner after bolus injection of [¹¹C]PIB tracer, and reconstructed with eight different settings⁴⁴ (data not shown). The results showed that (1) the variability of the metric is between 1% to 5%, depending on the reconstruction settings, (2) the mean relative difference between GMvsWM angles (i.e. angle between the kinetic vectors of the grey matter and white matter classes) computed with different reconstruction is 6% ± 6% (mean±SD) and (3) reproducibility of the angle measures (as intraclass correlation coefficient, ICC) was 0.79. This was measured in test-retest data using SPSS21 to compute the ICC using a two-factor mixed effects model and type consistency. Overall, the variability is too low to explain any of the differences between GMvsWM kinetic angles obtained in this study from different TSPO tracers.

Conclusion

Vascular binding is common to all TSPO tracers, regardless of their affinity, and it should be taken into account for tracer kinetic modelling to optimally account for the data. The magnitude of the vascular binding fraction is proportional to tracer affinity and also modulates the tracer contrast between grey and white matter. In particular, high-affinity tracers, such as [¹¹C]PBR28, which have relatively low tissue contrast, are poorly suited for supervised reference tissue analysis.

Footnotes

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: MV and FET were supported by the Wellcome Trust Strategic Award “Neuroinflammation in Alzheimer’s Disease and Depression”. MV is also supported by the National Institute for Health Research (NIHR) Biomedical Research Centre at South London and Maudsley NHS Foundation Trust and King’s College London. OH is supported by the Medical Research Council-UK (no. MC-A656-5QD30), Maudsley Charity (no. 666), Wellcome Trust (no. 094849/Z/10/Z), the National Institute for Health Research (NIHR) Biomedical Research Centre at South London and Maudsley NHS Foundation Trust and King’s College London. [¹⁸F]DPA714 acquisitions were supported by ARSEP and ANR MNP2008-007125 and sponsored by APHP.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Authors’ contributions

GR, MV, PZF, FET and AB made substantial contribution to the study design. GR, MV, MT, BB, BS, PSB, OH and FET made substantial contribution to the data collection and analysis. GR, MV, MT, BB, BS, CW, SL, MB, PSB, OH, PZF, FET and AB made substantial contribution to the data interpretation. GR, MV, MT, PZF, FET and AB drafted the manuscript. All the authors critically revised the article. All authors approved the last version.

Supplementary material

Supplementary material for this paper can be found at the journal website:

References

Liu

G-J

Middleton

Hatty

et al.

The 18 kDa translocator protein, microglia and neuroinflammation. Brain Pathol 2014; 24: 631–653.

Chen

M-K

Guilarte

. Translocator protein 18 kDa (TSPO): molecular sensor of brain injury and repair. Pharmacol Ther 2008; 118: 1–17.

Rupprecht

Papadopoulos

Rammes

et al.

Translocator protein (18 kDa) (TSPO) as a therapeutic target for neurological and psychiatric disorders. Nat Rev Drug Discov 2010; 9: 971–988.

Lavisse

Guillermier

Herard

A-S

et al.

Reactive astrocytes overexpress TSPO and are detected by TSPO positron emission tomography imaging. J Neurosci 2012; 32: 10809–10818.

Vivash

OBrien

. Imaging microglial activation with TSPO PET: lighting up neurologic diseases? J Nucl Med 2016; 57: 165–168.

Albrecht

Granziera

Hooker

et al.

In vivo imaging of human neuroinflammation. ACS Chem Neurosci 2016; 7: 470–483.

Hinz

Boellaard

. Challenges of quantification of TSPO in the human brain. Clin Transl Imaging 2015; 3: 403–416.

Turkheimer

Rizzo

Bloomfield

et al.

The methodology of TSPO imaging with positron emission tomography. Biochem Soc Trans 2015; 43: 586–592.

Owen

Guo

Rabiner

et al.

The impact of the rs6971 polymorphism in TSPO for quantification and study design. Clin Transl Imaging 2015; 3: 417–422.

10.

Martini

Chelli

Betti

et al.

Peripheral benzodiazepine binding sites in platelets of patients affected by mitochondrial diseases and large scale mitochondrial DNA rearrangements. Mol Med 2002; 8: 841–846.

11.

Soreni

Apter

Weizman

et al.

Decreased platelet peripheral-type benzodiazepine receptors in adolescent inpatients with repeated suicide attempts. Biol Psychiatry 1999; 46: 484–488.

12.

Pini

Martini

Abelli

et al.

Peripheral-type benzodiazepine receptor binding sites in platelets of patients with panic disorder associated to separation anxiety symptoms. Psychopharmacology (Berl) 2005; 181: 407–411.

13.

Begni

Tremolizzo

Andreoni

et al.

Neuroligand binding endophenotypes in blood cells distinguish two subsets of borderline personality disorder patients. Neurosci Lett 2009; 462: 144–146.

14.

Ritsner

Modai

Gibel

et al.

Decreased platelet peripheral-type benzodiazepine receptors in persistently violent schizophrenia patients. J Psychiatr Res 2003; 37: 549–556.

15.

Lockhart

Davis

Matthews

et al.

The peripheral benzodiazepine receptor ligand PK11195 binds with high affinity to the acute phase reactant alpha1-acid glycoprotein: implications for the use of the ligand as a CNS inflammatory marker. Nucl Med Biol 2003; 30: 199–206.

16.

Banati

Newcombe

Gunn

et al.

The peripheral benzodiazepine binding site in the brain in multiple sclerosis: quantitative in vivo imaging of microglia as a measure of disease activity. Brain 2000; 123: 2321–2337.

17.

Lyoo

Ikawa

Liow

J-S

et al.

Cerebellum can serve as a pseudo-reference region in Alzheimer disease to detect neuroinflammation measured with PET radioligand binding to translocator protein. J Nucl Med 2015; 56: 701–706.

18.

Veronese

Marques

Bloomfield

et al.

Kinetic modelling of [11C]PBR28 for 18 kDa Translocator Protein PET data: a validation study of vascular modelling in the brain using XBD173 and tissue analysis. J Cereb Blood Flow Metab Epub ahead of print 5 June 2017. DOI: 10.1177/0271678X17712388.

19.

Turkheimer

Edison

Pavese

et al.

Reference and target region modeling of [11C]-(R)-PK11195 brain studies. J Nucl Med 2007; 48: 158–167.

20.

Yaqub

van Berckel

BNM

Schuitemaker

et al.

Optimization of supervised cluster analysis for extracting reference tissue input curves in (R)-[(11)C]PK11195 brain PET studies. J Cereb Blood Flow Metab 2012; 32: 1600–1608.

21.

García-Lorenzo

Lavisse

Leroy

et al.

Validation of an automatic reference region extraction for the quantification of [18 F]DPA-714 in dynamic brain PET studies. J Cereb Blood Flow Metab Epub ahead of print 9 February 2017. DOI: 10.1177/0271678X17692599.

22.

Veenman

Gavish

. The peripheral-type benzodiazepine receptor and the cardiovascular system. Implications for drug development. Pharmacol Ther 2006; 110: 503–524.

23.

Cosenza‐Nashat

Zhao

Suh

et al.

Expression of the translocator protein of 18 kDa by microglia, macrophages and astrocytes based on immunohistochemical localization in abnormal human brain. Neuropathol Appl Neurobiol 2009; 35: 306–328.

24.

Bird

JLE

Izquierdo-Garcia

Davies

et al.

Evaluation of translocator protein quantification as a tool for characterising macrophage burden in human carotid atherosclerosis. Atherosclerosis 2010; 210: 388–391.

25.

Joo

Lee

Lim

et al.

Peripheral benzodiazepine receptor regulates vascular endothelial activations via suppression of the voltage-dependent anion channel-1. FEBS Lett 2012; 586: 1349–1355.

26.

Tomasi

Edison

Bertoldo

et al.

Novel reference region model reveals increased microglial and reduced vascular binding of 11C-(R)-PK11195 in patients with Alzheimer’s disease. J Nucl Med 2008; 49: 1249–1256.

27.

Rizzo

Veronese

Tonietto

et al.

Kinetic modeling without accounting for the vascular component impairs the quantification of [11C]PBR28 brain PET data. J Cereb Blood Flow Metab 2014; 34: 1060–1069.

28.

Bloomfield

Selvaraj

Veronese

et al.

Microglial activity in people at ultra high risk of psychosis and in schizophrenia: an [11 C]PBR28 PET brain imaging study. Am J Psychiatry 2016; 173: 44–52.

29.

Wimberley

Lavisse

Brulon

et al.

Impact of endothelial TSPO on the quantification of 18F-DPA714 in the normal human brain. J Nucl Med 2016; 57: 95.

30.

Spinks

Jones

Bloomfield

et al.

Physical characteristics of the ECAT EXACT3D positron tomograph. Phys Med Biol 2000; 45: 2601–2618.

31.

Kinahn

Rogers

. Analytic 3D image reconstruction using all detected events. IEEE Trans Nucl Sci 1989; 36: 964–968.

32.

Ranicar

Williams

Schnorr

et al.

The on-line monitoring of continuously withdrawn arterial blood during PET studies using a single BGO/photomultiplier assembly and non-stick tubing. Med Prog Technol 1991; 17: 259–264.

33.

Tonietto M, Rizzo G, Veronese M, et al. Modelling arterial input functions in positron emission tomography dynamic studies. In: 2015 37th annual international conference of the IEEE Engineering in Medicine and Biology Society (EMBC), Milan, Italy, 25–29 August 2015, pp.2247–2250. USA: IEEE.

34.

Cunningham

Jones

. Spectral analysis of dynamic PET studies. J Cereb Blood Flow Metab 1993; 13: 15–23.

35.

Hammers

Allom

Koepp

et al.

Three-dimensional maximum probability atlas of the human brain, with particular reference to the temporal lobe. Hum Brain Mapp 2003; 19: 224–247.

36.

Bertoldo

Vicini

Sambuceti

et al.

Evaluation of compartmental and spectral analysis models of [18F]FDG kinetics for heart and brain studies with PET. IEEE Trans Biomed Eng 1998; 45: 1429–1448.

37.

Innis

Cunningham

Delforge

et al.

Consensus nomenclature for in vivo imaging of reversibly binding radioligands. J Cereb Blood Flow Metab 2007; 27: 1533–1539.

38.

Akaike

. A new look at the statistical model identification. IEEE Trans Automat Contr 1974; 19: 716–723.

39.

Ory

Celen

Verbruggen

et al.

PET radioligands for in vivo visualization of neuroinflammation. Curr Pharm Des 2014; 20: 5897–5913.

40.

James

Fulton

Vercoullie

et al.

DPA-714, a new translocator protein-specific ligand: synthesis, radiofluorination, and pharmacologic characterization. J Nucl Med 2008; 49: 814–822.

41.

Banister

Beinat

Wilkinson

et al.

Ether analogues of DPA-714 with subnanomolar affinity for the translocator protein (TSPO). Eur J Med Chem 2015; 93: 392–400.

42.

Lavisse

Garcia-Lorenzo

Peyronneau

M-A

et al.

Optimized quantification of translocator protein radioligand 18F-DPA-714 uptake in the brain of genotyped healthy volunteers. J Nucl Med 2015; 56: 1048–1054.

43.

Delforge

Syrota

Bendriem

. Concept of reaction volume in the in vivo ligand-receptor model. J Nucl Med 1996; 37: 118–125.

44.

Veronese

Bodini

García-Lorenzo

et al.

Quantification of [11C]PIB PET for imaging myelin in the human brain: a test–retest reproducibility study in high-resolution research tomography. J Cereb Blood Flow Metab 2015; 35: 1–12.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.62 MB

Generalization of endothelial modelling of TSPO PET imaging: Considerations on tracer affinities

Abstract

Keywords

Introduction

Materials and methods

Vascular binding model for [11C]-R-PK11195

Dataset

PET imaging

Arterial blood sampling and analysis of [11C]-R-PK11195 plasma metabolism

Image analysis

[11C]-R-PK11195 kinetic analysis

Relationship between [11C]-R-PK11195, [18F]DPA714 and [11C]PBR28 vascular component and tracer affinity

The role of TSPO tracer affinity in using supervised-clustered reference region approaches

Results

[11C]-R-PK11195 PET kinetic analysis

Comparison of model performance

Kinetic estimates

Comparison of the vascular component for tracers with different affinity

Similarity of tissue classes kinetics

Discussion

TSPO vascular binding correction: Reference region or blood input function?

Vascular binding in TSPO tracers

Conclusion

Footnotes

Funding

Declaration of conflicting interests

Authors’ contributions

Supplementary material

References

Supplementary Material

Vascular binding model for [¹¹C]-R-PK11195

Arterial blood sampling and analysis of [¹¹C]-R-PK11195 plasma metabolism

[¹¹C]-R-PK11195 kinetic analysis

Relationship between [¹¹C]-R-PK11195, [¹⁸F]DPA714 and [¹¹C]PBR28 vascular component and tracer affinity

[¹¹C]-R-PK11195 PET kinetic analysis