Abstract
Neuroinflammation is associated with neurodegenerative disease. PET radioligands targeting the 18-kDa translocator protein (TSPO) have been used as in vivo markers of neuroinflammation, but there is an urgent need for novel probes with improved signal-to-noise ratio. Flutriciclamide (18F-GE180) is a recently developed third-generation TSPO ligand. In this first study, we evaluated the optimum scan duration and kinetic modeling strategies for 18F-GE180 PET in (older) healthy controls. Methods: Ten healthy controls, 6 TSPO high-affinity binders, and 4 mixed-affinity binders were recruited. All subjects underwent detailed neuropsychologic tests, MRI, and a 210-min 18F-GE180 dynamic PET/CT scan using metabolite-corrected arterial plasma input function. We evaluated 5 different kinetic models: irreversible and reversible 2-tissue-compartment models, a reversible 1-tissue model, and 2 models with an extra irreversible vascular compartment. The minimal scan duration was established using 210-min scan data. The feasibility of generating parametric maps was also investigated using graphical analysis. Results: 18F-GE180 concentration was higher in plasma than in whole blood during the entire scan duration. The volume of distribution (VT) was 0.17 in high-affinity binders and 0.12 in mixed-affinity binders using the kinetic model. The model that best represented brain 18F-GE180 kinetics across regions was the reversible 2-tissue-compartment model (2TCM4k), and 90 min resulted as the optimum scan length required to obtain stable estimates. Logan graphical analysis with arterial input function gave a VT highly consistent with VT in the kinetic model, which could be used for voxelwise analysis. Conclusion: We report for the first time, to our knowledge, the kinetic properties of the novel third-generation TSPO PET ligand 18F-GE180 in humans: 2TCM4k is the optimal method to quantify the brain uptake, 90 min is the optimal scan length, and the Logan approach could be used to generate parametric maps. Although these control subjects have shown relatively low VT, the methodology presented here forms the basis for quantification for future PET studies using 18F-GE180 in different pathologies.
Microglia play a crucial role as the first line of defense when neuronal damage occurs. Under normal conditions, microglial cells are in a state of surveillance with elongated processes. During injury or neurodegenerative processes, microglial cells become activated; the activated state is mirrored by upregulation of translocator protein (TSPO) (1,2). PET with TSPO-specific ligands provides an in vivo technique to detect microglial activation (3).
11C-(R)-PK11195 PET has been used for more than 2 decades. However, because of a low signal-to-background ratio and short half-life (20 min), second-generation TSPO markers have been developed. These include 18F-FEPPA, 18F-FEDAA1106, 11C-vinpocetine, 11C-DAC, 11C-DAA1106, 11C-N1-methyl-2-phenylindol-3-ylglyoxylamide, 11C-CLINME, 11C-DPA-713, 18F-DPA-714, 18F-PBR06, and 11C-PBR28 (4). However, the results obtained by second-generation radioligands in preclinical models have not been consistently reproduced in humans (5). Their quantification in humans suffers from 3 confounding factors (6). The first factor is genetic: a single nucleotide polymorphism in the TSPO gene (rs6971) leads to an amino-acid substitution (A147T) and reduced binding affinity. The second factor is the disproportion between the high signal from the TSPO in the endothelial cells of the blood–brain barrier (BBB) and venous sinuses and the signal from the tissue, requiring appropriate kinetic correction (7). The third factor is the difficulty in obtaining accurate estimates of free plasma concentrations for proper quantification.
Flutriciclamide (18F-GE180) has been identified as a promising TSPO tracer in preclinical models of stroke and lipopolysaccharide-induced central nervous system inflammation (8,9). Here we report the results of a study evaluating 18F-GE180 PET in healthy human brains. We have investigated blood tracer concentrations, brain uptake, quantification of binding via tracer kinetic modeling, optimal scan length, and feasibility of generating parametric maps.
MATERIALS AND METHODS
Fifteen healthy controls (HCs) aged 50–85 y were recruited. Subjects who were high-affinity binders (HABs = 6) or mixed-affinity binders (MABs = 4) proceeded to MRI and PET scans, whereas low-affinity binders (n = 5) were excluded. Details of recruitment, demography, and MRI acquisition are provided in the supplemental materials (available at http://jnm.snmjournals.org).
GE180 PET
18F-GE180 was manufactured on the FASTlab (GE Healthcare) (10). All subjects received 185 MBq of 18F-GE180 by bolus intravenous injection (in 20 s) immediately before the PET scan. The scan was acquired using a Biograph 6 PET/CT scanner (Siemens) (11). An initial CT was acquired for patient position and for attenuation correction of PET images. The tracer was then injected, and dynamic emission PET images were acquired over 210 min using predetermined time frames: 6 × 15, 3 × 60, 5 × 120, 5 × 300, and 14 × 600 s. Images were corrected for attenuation, random, and scattered emissions based on the 3-dimensional ordinary Poisson ordered-subset expectation maximization algorithm. Reconstruction of PET was performed using filtered backprojection.
Blood Data
Arterial whole-blood activity was continuously monitored for the first 15 min of the PET scan using the Allogg automated blood sampling system with real-time online blood sampling. Ten discrete arterial blood samples were taken at 5, 10, 15, 30, 60, 90, 120, 150, 180, and 210 min of the PET scan. The activity concentration was measured in both whole blood and plasma by a well-counter and was used to generate a plasma-to-blood ratio curve. The time course of activity in the plasma can be estimated with a sigmoidal fitting using the plasma-to-blood ratio model to the first 15 min of continuous whole-blood data.
Metabolites of 18F-GE180 were measured by high-performance liquid chromatography analysis (10) using discrete blood samples. A 2-exponential linear model was used to describe the parent fraction of 18F-GE180, which was applied to the plasma to generate the parent plasma input function. SUV was applied to both total blood and parent plasma with the formula:
Finally, the time delay (the arrival of the 18F-GE180 bolus at the peripheral sampling site) was determined using MICK.exe (quantitative software for the PET analysis) and was performed using MATLAB 2014 (The MathWorks, USA) on a Windows platform.
Region-of-Interest (ROI) Analysis
Each subject underwent two 90-min PET scans with a 20-min break in between; the realignment approach was necessary to realign the second scan to the first scan. Head motions were corrected using the frame-by-frame realignment tools in statistical parametric mapping. On the basis of the visual resolution, the eighth frame (185–245 s) was selected as the reference frame for each time frame realigning to. An individualized object map in PET space was created with the following procedure: MRI was coregistered to the PET add image (60–90 min), a binary gray and white matter mask was created via segmentation, the probabilistic ROI atlas (12) in MNI space was transformed into native PET space, and a binary mask was applied to the ROI atlas to generate individual ROIs. Regional time–activity curves were generated by sampling the dynamic PET images with the individual ROIs for the following merging regions: frontal, temporal, parietal, and occipital lobes and whole brain. Additional sampling of the posterior cingulate, thalamus, brain stem, whole medial temporal lobe, hippocampus, and cerebellum was performed. The parahippocampus, anterior cingulate, and amygdala were sampled as additional ROIs using Analyze 11.0 (AnalyzeDirect). The SUV time–activity curve was generated for each 18F-GE180 dynamic PET by dividing the ratio of 18F-GE180–injected dose over body weight.
Kinetic Modeling
Parent plasma input function and dynamic PET were used to investigate the best kinetic model to compute 18F-GE180 total volume of distribution (VT) from time–activity curves with MICK software (Modeling-Input-function-Compartmental-Kinetics) and MATLAB. We evaluated 5 different kinetic models (supplemental materials). 18F-GE180 tissue data were investigated with the reversible 1-tissue 2k (1TCM2k) model, the 2-tissue 4k (2TCM4k) model, and the irreversible 2-tissue 3k, for which k implies the rate constant for tracer for different kinetic compartments. As tracer binds to BBB endothelial cells, endothelia could significantly affect kinetic modeling (7,13); therefore, we also considered 1 tissue with extra vascular component (1TCM2k-1k) and 2-tissue with extra vascular component (2TCM4k-1k). The additional component describes the trapping of the tracer by the endothelial cells of blood vessels. The Akaike information criterion (AIC) (14) is a statistical measure to estimate the quality of fit of the predicted model against the actual data using different statistical models. In simple terms, the AIC values indicate the information lost when a candidate model is applied to estimate the real data, and the model with the smallest AIC value should be selected as the preferred model, as it indicates the best fit (15). To select the best model to quantify the tracer, AIC is calculated using the following formula:
where m denotes number of parameters, n equals the sum of degrees of freedom and number of parameters, and wrss denotes weighted residual sum of squares (16). The AIC fraction was also calculated by measuring the frequency of AIC preferences for each model across all subjects (17). AIC was calculated for the following brain regions: frontal, temporal, parietal, and occipital lobes; whole brain; posterior cingulate; thalamus; striatum; brain stem; medial temporal lobe; hippocampus; and cerebellum. Coefficient of variation (CV) was measured to assess dispersion and precision of parameters. We also evaluated the separation in average binding between HAB and MAB classes.
Graphical Analysis
Graphical analysis was applied to investigate the feasibility of generating parametric maps of 18F-GE180. We used the Logan plot to generate pixel-wise parametric maps. Initially we calculated the optimal threshold time by linear fit, the tissue-to-plasma ratio against the time. 18F-GE180 parametric maps were generated with the parent plasma input function and dynamic PET images using MICK parametric map software on MATLAB.
Optimization of Scan Length
The optimum scan duration was evaluated, measuring the percentage change of k (rate constants) and VT, along with their CV at different time durations, calculated at 60, 75, 90, 150, 180, and 210 min. k3 and k4 indicate the rate constant of 18F-GE180 in and out of the specific-bound compartment, whereas the k3/k4 ratio represents the specific binding of the tracer. Thus, the k3/k4 ratio was measured to investigate whether the binding potential of 18F-GE180 was affected when the scanning time was reduced.
Statistical Analysis
Statistical analysis of groupwise differences between different kinetic models, different scan durations, and ROI VT in different genetic groups was calculated in SPSS for Windows (version 22; SPSS). A P value of less than 0.05 was regarded as a statistical difference. The Pearson correlation was applied to measure the linear correlation between 2 groups of variables using SPSS. Repeated-measures ANOVA was applied to detect any overall differences between repeated measurements.
RESULTS
Parent Plasma Input
Consistent with the preclinical work (8), the radiochromatogram revealed 3 identifiable radiolabeled metabolites and a parent 18F-GE180 in human plasma (Fig. 1A). The parent fraction of 18F-GE180 gradually reduced from 90% to 60%. There was no group difference between HABs and MABs in metabolic rates and plasma-to-blood curve (Figs. 1B and 1C). The 18F-GE180 activity concentration in plasma was 1.6-fold higher than in whole blood. No difference in parent plasma input function between HABs and MABs was found (Fig.1D).
18F-GE180 blood data. (A) HPLC analysis for 18F-GE180 at 15 min in HAB subject (solid line) and MAB subject (dashed line). (B) 18F-GE180 parent fraction in arterial plasma. (C) Plasma-to-blood ratio in HABs (●) and MABs (▪). (D) SUV for parent plasma and whole blood as group and parent plasma input function curve in either HAB or MAB cohorts.
Dynamic PET Image Time–Activity Curve
Because tracer concentrations are correlated with injected dose and individual weight, we measured SUV time–activity curve values at each time frame. The tracer activity concentration in the brain peaked at 125 s, with SUV at 1.5, and then quickly washed out and reached a plateau around 40 min. As a group, SUVs were significantly higher in HABs than MABs in the temporal lobe (P = 0.0001), thalamus (P = 0.0001), and striatum (P = 0.0002) (Fig. 2). The venous sinuses showed consistently highest tracer activity throughout the scan, whereas gray matter showed around 30% higher activity concentration than white matter.
SUV corrected time–activity curves of 18F-GE180 in temporal and occipital lobes (A and B), thalamus (C), and striatum (D) for HABs (○) and MABs (*).
Kinetic Modeling
Among 5 kinetic models, 2TCM4k showed the lowest AIC values in frontal (−59.6 ± 5.4), temporal (−46.4 ± 4.5), parietal (−58.6 ± 5.4), and occipital lobes (−54.3 ± 5.1) as a group (Fig. 3A). The 2TCM4k model also showed the highest AIC fraction (Fig. 3B). Although the 2TCM3k irreversible model demonstrated closer AIC values (<10% difference compared with the 2TCM4k model), the Ki (net influx constant), demonstrating the overall net rate of tracer uptake into tissue in the 2TCM3k model, showed poor precision with a high CV (104% ± 85%) whereas VT in the reversible 2TCM4k model had a CV of 21% ± 13%. On the basis of the model curve fitting, 2TCM4k provided a good fit to 18F-GE180 activity, supported by smallest weighted residual sum square and more random in residual’s sequence (similar number of residuals above or below the model curve) using the Wald–Wolfowitz test. In summary, the 2TCM4k is the preferred method of analysis of 18F-GE180 PET. The supplemental materials detail rate constants (K1, k2, k3, and k4), blood volume, and VT, applying the 2TCM4k model in HABs and MABs, which showed good precision of parameter estimation.
Kinetic model selection. (A) AIC values for 5 models in frontal, temporal, parietal, and occipital lobes and whole brain. (B) Graph with AIC fraction, revealing frequency of AIC preferences for each model across all subjects.
VT
As a group, 10 HCs had low mean VT in the whole brain but slightly higher in temporal the lobe, occipital lobe, and thalamus (Fig. 4). The thalamus, parahippocampus (47%, P = 0.04), and cerebellum showed significantly higher mean VT values in HABs than in MABs (Table 1).
18F-GE180 regional VT in 2TCM4k model. Regional VT with application of 2TCM4k model in 11 ROI regions. Circle with black bar represents HABs, and square with white bar represents MABs. M = medial; P = posterior.
Regional VT for 18F-GE180 Applied 2TCM4k Model in All HCs, HABs, and MABs
Scan Length
The 210-min data were excluded because variance was high toward the end of the scan. Repeated ANOVA did not show difference in the absolute VT when the scan duration was reduced. In this study, the group mean of k3/k4 ratio (HAB = 0.52 and MAB = 0.34) was calculated to evaluate the results among different scan lengths, showing no differences with scan length reduction from 180 to 90 min. When the scan length was further reduced to 75 min, the k3/k4 ratio remained the same as with the 90-min data, but 4 individual subjects showed a 14% reduction in k3/k4 ratio. Compared with 90-min scans, 60-min scans showed a 51% reduction in k3/k4 ratio as a group. These data suggest that for this cohort the optimal scan duration is 90 min.
Logan
Kinetic modeling suggested that 18F-GE180 has a reversible feature, therefore we applied Logan graphical analysis (18) to generate parametric maps. On the basis of linear fit of the whole brain and the appropriate weight for different time points, dynamic data from 20 to 90 min (8 data points) were selected to generate a Logan straight line (slope = 0.147 ± 0.06 and intercept = −1,623 ± 387) for parametric maps (Fig. 5). The Logan VT values (supplemental materials) were correlated with 2TCM4k VT values, with a strong correlation in 4 cortices and hippocampus in both HABs (P < 0.0001) and MABs (P < 0.0001) (Fig. 6). The HABs showed generally higher Logan VT values than MABs; however, there were no statistically significant differences in this small sample.
Individual 18F-GE180 Logan parametric maps in transverse view are demonstrated with corresponding MR image.
VT in 2-tissue-reversible model and Logan parametric map demonstrate similar VT results with strong Pearson correlation in 4 cortices and hippocampus in both HABs (P < 0.0001) and MABs (P < 0.0001).
DISCUSSION
To our knowledge, this is the first study evaluating different approaches for PET modeling of the third-generation TSPO tracer 18F-GE180 in older healthy subjects, establishing the optimum scanning time in this study population. There was consistently higher 18F-GE180 activity in plasma than in whole blood, but brain VT values generated from plasma input functions were low (∼0.16). Data were acquired for 210 min after injection of 18F-GE180, and 5 different kinetic models were evaluated in this study. The 18F-GE180 VT remained unchanged when the scan length was reduced to 90 min, but changes were observed when the scan length was reduced to 75 and 60 min, suggesting that 90 min might be the minimal scan time for 18F-GE180 PET. The 2TCM4k model demonstrated the lowest AIC and CV among the 5 kinetic models, suggesting that the 2TCM4k model could be the best model to estimate 18F-GE180 uptake. Compared with the 2TCM4k model, Logan graphical analysis allows us to create parametric maps and to estimate VT at the voxel level. Logan VT was highly correlated with 2TCM4k VT in all cortices, indicating that the Logan graphical analysis can be used to create unbiased parametric VT maps for 18F-GE180.
Consistent with a preclinical study (8), high-performance liquid chromatography revealed 3 separate 18F-GE180 plasma metabolites, with a similar profile in HAB and MAB and no significant differences between HABs and MABs in plasma-to-blood ratios. The plasma-to-blood ratio for tracers crossing the BBB by passive diffusion is close to unity at the beginning of PET. 18F-GE180 showed a relatively high plasma-to-blood ratio of 1.55, indicating higher concentration in plasma than red blood cells. A persistently high plasma-to-blood ratio throughout the scan reflects poor membrane penetration of 18F-GE180 across red cell membranes. In vitro assays of plasma protein binding of 18F-GE180 in human samples showed a relatively high percentage (97.4%) plasma protein binding, which was also seen with other TSPO tracers.
Preclinical animal models showed that 18F-GE180 crosses the BBB readily. In humans, 18F-GE180 demonstrated a relatively low VT in the brain, possibly because of low availability of the GE180 due to significant plasma protein binding, low BBB permeability, or clearance by efflux pumps. Despite the fact that species differences have been observed in the BBB structure (19) and other novel brain PET tracers (20), the significant difference observed between human and preclinical models in 18F-GE180 could be primarily due to a significantly higher binding to plasma protein in humans.
18F-GE180 preclinical studies demonstrated that, in rats, there is a clear regional differentiation of high-TSPO-expressing regions (olfactory bulbs) compared with low-TSPO-expressing regions (striatum) (21). In this study, 5 subjects (1 MAB and 4 HABs) revealed lowest uptake in the striatum: 3 (MABs) showed lowest uptake in the cerebellum and 2 (HABs) showed a more generalized distribution of TSPO expression. This inconsistent pattern suggests that a single predefined reference region for nonspecific binding may not be appropriate for 18F-GE180 in humans in this age group.
Because TSPO expression is low in the healthy brain, the 18F-GE180 VT values were relatively small in all ROI regions (range, 0.15–0.20). A higher signal in the thalamus than other regions was found, consistent with previous 11C-(R)-PK11195 studies (22). However, VT values for 18F-GE180 suggest that tracer brain uptake is relatively low compared with other TSPO PET radioligands, though the impact of this radioligand as a microglial imaging marker cannot be determined until we have results in specific disease conditions. The significant inverse correlation between the 18F-GE180 input function and the tracer uptake in the brain reflects the higher uptake to plasma proteins in humans than animal models.
Interestingly, the significant inverse correlation between 18F-GE180 input function and VT in the brain for each binding class separately, with a correlation coefficient greater than 0.9, indicates that whole-brain VT can be almost entirely predicted by plasma levels within each binding class. As we were unable to estimate plasma free fractions reliably, the fluctuations in estimated VT may be due to fluctuations in free plasma fractions across subjects. Indeed, higher plasma protein binding would result in higher plasma counts and lower VT. This observation replicates similar observations in other TSPO tracers (13,23). Identification of an optimal reference method would allow us to further refine the analysis methods without the arterial input.
Most second-generation TSPO tracers, such as 11C-PBR28 and 11C-PBR111, demonstrated an association with the TSPO genotype, revealing a 30% higher uptake in the whole brain of HABs than MABs. In this study, there was a 30%–40% higher uptake in HABs, reaching significance in the thalamus, parahippocampus, and cerebellum. Nevertheless, there was some overlap between subjects; moreover, the study cohort was also small. Compared with some second-generation TSPO tracers, 18F-GE180 may be less sensitive to the TSPO binding affinity status and may behave more like 11C-(R)-PK11195. One could argue, some of the variability seen in the MABs and HABs could be due to test–retest variability, which we are evaluating. The high correlation between Logan VT and 2TCM4k VT suggests that Logan parametric mapping could be used to generate pixelwise parametric VT maps. Because Logan generally underestimates the VT (24), there was no statistically significant difference between HAB and MAB, potentially because of the small numbers in this pilot study. Additionally, consistently with VT results, the Logan VT parametric maps demonstrated heterogeneous distribution of 18F-GE180 uptake in the HCs. This heterogeneity was observed in both HAB and MAB subgroups and was consistent with 2TCM4k model.
Although one of the limitations of this study is the small sample size, this pilot study provides the first insight into the performance of 18F-GE180 in the normal human brain and information about the parent plasma input function. Another limitation is the exclusion of the low-affinity binders from the study, and our laboratory is recruiting more participants to evaluate this tracer further. The low expression of TSPO in the normal brain could be another limitation of this study evaluating only HCs. However, these scans were obtained with an arterial input function and an extensive scanning time of 210 min, providing significant information about the tracer kinetics and modeling of this novel TSPO tracer for future use.
CONCLUSION
This pilot study provides preliminary brain uptake kinetics of the novel TSPO tracer 18F-GE180 in HCs. We have demonstrated for the first time, to our knowledge, that 18F-GE180 shows uptake into the brain of HCs and that the tracer VT could be effectively modeled using 2TCM4k and Logan graphical models to quantify the TSPO binding in the human brain. The plasma-to-blood ratio and the rate of metabolism and input functions were not different between HAB and MAB subjects. We have also demonstrated that 90 min is the optimal scan duration, providing enough information to generate the precise and accurate estimates of VT. Semiquantitative SUV analysis for tracer uptake demonstrated a good correlation with kinetic modeling VT values. Logan graphical analysis with arterial input function enables us to generate 18F-GE180 parametric maps for voxelwise analyses.
DISCLOSURE
The costs of publication of this article were defrayed in part by the payment of page charges. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734. This work received financial support from Alzheimer’s Research U.K., GE Healthcare, and the National Institute for Health Research. Dr. Paul Edison was funded by the Higher Education Funding Council for England and received grants from Alzheimer’s Research, U.K., Alzheimer’s Drug Discovery Foundation, Alzheimer’s Society, U.K., Novo Nordisk, Piramal Life Science, and GE Healthcare. Calsolaro Valeria was funded by the Alzheimer’s Research U.K. fellowship. Drs. William Trigg and Christopher Buckley were funded by GE Healthcare. Prof. David J. Brooks received research grants from the Medical Research Council and Alzheimer's Research Trust, during the conduct of the study, as well as other grants from GE Healthcare; personal fees from AstraZeneca, Cytox, Shire, Novartis, GSK (Holland), Navidea, UCB, and Acadia; and grants from the Michael J. Fox Foundation and European Commission, outside the submitted work. No other potential conflict of interest relevant to this article was reported.
Acknowledgments
We thank GE Healthcare for providing the radiotracers and the Imperial College Clinical Imaging Facility for providing MRI and PET imaging facilities. We thank Dr. Albert Busza, Andy Blyth, Stephanie McDevitt, Neva Patel, and Gokul Kolipaka for 18F-GE180 scanning.
Footnotes
↵* Contributed equally to this work.
Published online Jun. 3, 2016.
- © 2016 by the Society of Nuclear Medicine and Molecular Imaging, Inc.
REFERENCES
- Received for publication November 23, 2015.
- Accepted for publication May 10, 2016.
Jump to section
Related Articles
Cited By...
- Quantification of Macrophage-Driven Inflammation During Myocardial Infarction with 18F-LW223, a Novel TSPO Radiotracer with Binding Independent of the rs6971 Human Polymorphism
- Head-to-Head Comparison of 11C-PBR28 and 18F-GE180 for Quantification of the Translocator Protein in the Human Brain
- Impact of Endothelial 18-kDa Translocator Protein on the Quantification of 18F-DPA-714