- Original research
- Open Access
Test-retest reliability and convergent validity of (R)-[11C]PK11195 outcome measures without arterial input function
EJNMMI Research volume 8, Article number: 102 (2018)
The PET radioligand (R)-[11C]PK11195 is used to quantify the 18-kDa translocator protein (TSPO), a marker for glial activation. Since there is no brain region devoid of TSPO, an arterial input function (AIF) is ideally required for quantification of binding. However, obtaining an AIF is experimentally demanding, is sometimes uncomfortable for participants, and can introduce additional measurement error during quantification. The objective of this study was to perform an evaluation of the test-retest reliability and convergent validity of techniques used for quantifying (R)-[11C]PK11195 binding without an AIF in clinical studies.
Data from six healthy individuals who participated in two PET examinations, 6 weeks apart, were analyzed. Regional non-displaceable binding potential (BPND) values were calculated using the simplified reference tissue model, with either cerebellum as reference region or a reference input derived using supervised cluster analysis (SVCA). Standardized uptake values (SUVs) were estimated for the time interval of 40–60 min.
Test-retest reliability for BPND estimates were poor (80% of ICCs < 0.5). BPND estimates derived without an AIF were not correlated with BPND, total or specific distribution volume from the 2TCM using an AIF (all R2 < 12%). SUVs showed moderate reliability but no correlation to any other outcome measure.
Caution is warranted when interpreting patient-control comparisons employing (R)-[11C]PK11195 outcome measures obtained without an AIF.
(R)-[11C]PK11195 was the first positron emission tomography (PET) radioligand developed for quantification of the translocator protein (TSPO). Within the brain, TSPO is mainly expressed in glial cells. Based on in vitro studies showing increases in TSPO expression in response to pro-inflammatory stimuli, the protein has been considered a biomarker for brain immune activation . As such, (R)-[11C]PK11195 has, since the early 1990s, been applied in a wide range of clinical studies .
TSPO is expressed throughout the brain which means that no region can serve as a reference for quantification of specific (R)-[11C]PK11195 binding. Instead, a metabolite-corrected arterial input function (AIF) must be obtained and used as an input function for a kinetic model from which binding parameters can be estimated. Common measures of regional binding derived from the use of an AIF are total distribution volume (VT), specific distribution volume (VS or BPP), and binding potential (BPND) .
Obtaining a metabolite-corrected input function is costly, often uncomfortable for research participants, and can also be prone to measurement error. Therefore, alternative approaches for quantifying binding have been suggested which are less demanding and which do not require an AIF. The most simple method is to calculate the radioactivity concentration in a brain region normalized by the injected radioactivity and the subject’s weight (standardized uptake value or SUV). As such, the SUV does not directly reflect specific binding since the signal also contains non-specific binding and radioactivity from vasculature. Importantly, SUVs are also dependent on the rate and extent of radioligand delivery to the brain. This means that results may be influenced by cerebral blood flow or peripheral changes such as differences in metabolism or blood binding. For TSPO, previous studies have shown a high correlation between (R)-[11C]PK11195 VT and SUVs in synovial tissue in the knee joint [4, 5]. However, it has been shown that in pathological conditions characterized by reduced rates of flow or variations in blood volume, the use of SUVs for quantifying (R)-[11C]PK11195 brain uptake yields biased and unprecise outcomes . In addition, the high concentration of TSPO in peripheral tissues, which may be altered during peripheral inflammation, can greatly influence the extent of radioligand brain delivery . Due to these reasons, SUVs are unlikely to be an unbiased index of TSPO binding in brain.
An alternative method for quantifying (R)-[11C]PK11195 binding without the use of an AIF is the supervised cluster analysis (SVCA) method [8, 9]. SVCA, which is performed on dynamic PET images, aims to segment voxels into classes, differentiated by their kinetic behavior. The goal is to isolate gray matter (GM) voxels assumed to contain negligible levels of specific binding. These voxels are then used to establish a time-activity curve (TAC) serving as a reference input in a kinetic model, such as the simplified reference tissue model (SRTM) . However, application of the SVCA method has some limitations. The kinetic classes can produce different results depending on scanner type, acquisition protocol, and radioligand, and it is important to evaluate the classes prior to applying them clinically. The SVCA method has been used in (R)-[11C]PK11195 studies to compare TSPO binding between healthy control subjects and patients with Alzheimer’s disease [11, 12], multiple sclerosis , traumatic brain injury , and schizophrenia  or studies which examined changes in TSPO expression in normal aging [16, 17].
Another simplified approach to obtain BPND values without arterial sampling is to use a reference tissue model with the cerebellum as reference region, despite the fact that the cerebellum contains non-negligible levels of TSPO . Since there is specific binding of (R)-[11C]PK11195 in the reference region, ensuing BPND values will not reflect the “true” binding potential but rather pseudo-binding potential (pseudo-BPND). One previous study found that pseudo-BPND was correlated to BPND from a constrained version of the 2TCM, using a sample consisting of elderly and patients with traumatic brain injury . Pseudo-BPND values have also been used, for example, to compare (R)-[11C]PK11195 binding in healthy controls and in patients with schizophrenia [20, 21], major depressive disorder , and glioma .
In order for PET quantification methods to be useful in clinical studies, they should yield outcomes which are both reliable and valid. Two previous studies evaluated a set of different methods, with and without AIF, for quantifying (R)-[11C]PK11195 binding [19, 24], but did not examine the test-retest reliability of ensuing outcome measures. In a previous test-retest study of six healthy subjects performed at our center, the reliability of (R)-[11C]PK11195 BPND values obtained using AIF were found to be very poor in most target regions examined . In contrast, the test-retest reliability of (R)-[11C]PK11195 BPND using SRTM with SVCA reference has been evaluated in four patients with Alzheimer’s disease . In that study, ICC values were found to be high in most regions of interest. However, no study has yet examined the test-retest reliability of SVCA in healthy controls. To our knowledge, the reliability of (R)-[11C]PK11195 SUV or BPND from SRTM with the cerebellum as reference has never been reported in healthy control subjects, despite both outcomes being applied in clinical patient-control comparisons.
The main objective of this study was to evaluate the test-retest reliability and repeatability of (R)-[11C]PK11195 (1) SUVs and (2) BPND obtained from SRTM, using cerebellum or SVCA-derived voxels as reference, respectively. The second objective was to examine the convergent validity of these outcomes by correlating them to VT, VS, and BPND values derived using an AIF.
Methods and materials
Subjects and imaging procedures
In the present analysis, we included PET examinations from six healthy male subjects (mean age = 25.8±3.9) who participated in a previous test-retest study of (R)-[11C]PK11195 . All subjects gave written informed consent according to the Helsinki Declaration prior to their participation in the original study. The study was approved by the Karolinska University Hospital Radiation Safety Committee and the Regional Ethics Committee in Stockholm.
All subjects participated in two PET measurements that took place approximately 6 weeks apart and were run on an ECAT Exact HR 47 system (Siemens/CTI, Knoxville, TN, USA). Structural magnetic resonance imaging (MRI) examinations were performed on a Siemens 1.5T Magnetom, resulting in a T1-weighted image for each subject. Production and radio-synthesis of (R)-[11C]PK11195 has been described previously . Mean injected radioactivity was 302 ± 33 MBq. Arterial samples were obtained in all PET measurements, from which a metabolite-corrected AIF was derived (see ).
ROI delineation was performed on the subjects’ T1-weighted images using the FreeSurfer software (5.0.0, http://surfer.nmr.mgh.harvard.edu/). ROIs were co-registered to PET images using SPM5 (Wellcome Department of Cognitive Neurology, UK). Sixty-three-minute TACs were extracted for the whole of GM, frontal cortex, striatum, thalamus, hippocampus, and cerebellum (CER), except for one PET examination where only a 50-min scan was obtained.
Quantification of outcomes with and without AIF
The two-tissue compartment model (2TCM) with AIF was used to estimate kinetic rate constants. The fraction of blood volume in target tissue (vB) and the delay between start of the AIF and the ROI TAC were fitted using the 2TCM applied on the entire GM TAC. These parameters were held constant for the remaining ROI fits. VT, VS, and BPND were then calculated using the rate constants. In addition to the above, we also evaluated the outcomes from the 2TCM when fitting vB separately for each ROI, as this has been suggested to yield less bias and better fit for BPND (k3/k4) estimates .
SUVs were calculated from the average radioactivity concentration in frames spanning from 40 to 60 min of the regional TACs and dividing by the injected radioactivity and the subject’s body weight. A time span of 40–60 min was chosen since this has previously shown to produce SUVs which were associated with VT in knee joints in patients with rheumatoid arthritis . However, we also evaluated three additional time intervals spanning from 10 to 30, 20 to 40, and 30 to 50 min.
The original SVCA method classifies PET voxels into six different tissue types associated with distinct kinetic profiles: (1) GM with high specific binding, (2) GM with low specific binding, (3) white matter, (4) soft tissue, (5) bone, and (6) blood. It has been shown that removal of bone and soft tissue, by using a MRI defined brain mask, prior to performing SVCA reduced variability of binding estimates and improved correlation to outcomes derived using an AIF . We therefore applied this restricted SVCA method (SVCA4), using the MATLAB software “Super-PK” (Imperial Innovations, Imperial College London). The Super-PK software was modified in order to be compatible with the scanning protocol applied in this study. Specifically, a cubic Gaussian smoothing kernel (FWHM 4 mm) was applied to all PET images prior to the analysis, and the 30-s background frame present in the population-based kinetic classes was removed. A reference TAC was then obtained for each PET measurement consisting of GM voxels classified as being associated with low specific binding. SRTM (called SRTM-SVCA4 below) was applied to estimate BPND for all ROIs. In this study, the primary results from SVCA4 method is based on the kinetic classes from Turkheimer et al. . However, we also evaluated the reliability of outcomes derived using two additional sets of population-based kinetic classes: one set from VU University Medical Center Amsterdam (VUMC)  and an unpublished set from the Turku PET Centre (TPC). This was done in order to examine the robustness of the SVCA4 method when using different population-based kinetic classes.
It has been shown that by using a version of SRTM that takes the radioactivity contribution from the vasculature into account, separation in (R)-[11C]PK11195 SVCA4 derived BPND between patients with AD and healthy controls can be improved [9, 27]. In addition to the SRTM algorithm, this model (called SRTMv) estimates and corrects for the fraction of blood volume in both target and reference TACs, by using an image-derived blood curve. Hence, we also evaluated the performance of SRTMv when using a reference curve derived from SVCA4 (SRTMv-SVCA4). Image-derived blood curves were obtained by extracting radioactivity from the entire scan from a region defined by the 10 voxels of highest intensity from the first minute of each examination, as described previously .
The SRTM with the cerebellum as pseudo-reference region (SRTM-CER) was also applied on all PET measurements and TACs to obtain BPND values for each ROI.
Finally, we also calculated VT, VS, and BPND values from the cerebellum and the SVCA4 reference TAC using the 2TCM with an AIF. This was done in order to both ascertain that the results were similar to previously published data and to evaluate the reliability and stability of the reference input.
The test-retest reliability, repeatability, and precision were examined by calculation of the intraclass correlation coefficient (ICC), the percentage average absolute variability (AbsVar), and the standard error of measurement (SEM), respectively. Since AbsVar can scale with the additive magnitude of the outcome, this particular metric is not suitable for comparing different outcomes with different means. We therefore also report the test-retest metric minimum detectable difference (MD). MD is based on the precision of an outcome (SEM) and is an approximation of the size of a difference from one measurement to another measurement which would be needed to detect a “real” change (according to a 95% confidence interval; ). MD is reported as a percentage of the absolute mean of the outcome, in order to allow for comparison between different measures.
Convergent validity was examined by correlating all outcomes without an AIF to those derived using an AIF. Although the outcomes are derived in different ways, they all aim to estimate, directly or indirectly, the specific binding to target. If they are valid outcomes, they should therefore be correlated to one other. For instance, both VS and BPND are defined as being proportional to the availability of target (∝ Bavail/Kd), and a lack of correlation would therefore imply that one or both of the outcomes are expressing a high degree of error or imprecision. Another reason for evaluating the correlations between outcomes is to inform the design of future meta-analyses. Outcomes which are only weakly associated with one another should likely not be entered into the same meta-analytic model, as this would violate a critical statistical model assumption and thereby yield uninterpretable estimates of effect size.
All kinetic modeling were performed using the R-package “kinfitr” (version 0.3.0, www.github.com/mathesong/kinfitr) together with “nls.multstart” . All statistical analyses were carried out in R (v.3.3.2 “Sincere Pumpkin Patch”).
Test-retest reliability of outcome measures derived with and without an AIF
Table 1 shows the mean, SD, and test-retest metrics for all outcomes. BPND values from SRTM using SVCA4 and the cerebellum as reference, and SRTMv using SVCA4 as reference were in the same range as described previously for healthy control subjects [8, 9]. There was a large difference in magnitude of BPND values derived with and without the use of an AIF. Regional BPND values from the 2TCM were on average 7 times higher than BPND from SRTM-SVCA4 and over 700 times higher than BPND from SRTM-CER. VT values for the SVCA4 reference TACs were of a similar magnitude and range (mean = 0.74, SD = 0.18, range = 0.49 to 0.96) compared to previously published results .
In the present analysis, SUVs, VT, and VS had the highest reliability across all ROIs (median ICCSUV = 0.84; median ICCVT = 0.69; median ICCVS = 0.67). BPND from SRTM and SRTMv with SVCA4 reference showed the lowest overall reliability (median ICC = 0.21 and − 0.14).
SUV, VT, and VS showed on the smallest (and therefore best) minimum detectable difference (median MDSUV = 38; median MDVT = 43; median MDVS = 34), while BPND from SRTM-CER showed the highest MD (median MD = 444).
Test-retest reliability of binding in the cerebellum and the SVCA4-derived reference
Figure 1 displays the average of all subjects’ TACs for the cerebellum ROI, SVCA reference, and the metabolite-corrected plasma curve, as well as the thalamus ROI.
Using the 2TCM, we derived VT, VS, and BPND (k3/k4) values for the cerebellum ROIs, as well as for the SVCA4 reference TACs. ICC, AbsVar, and MD values were similar to those reported for the target ROIs (Table 2). VT values were in the same range as described previously (see Fig. 3 in ).
Evaluation of additional population-based kinetic classes for supervised cluster analysis
In addition to the population-based kinetic classes developed by , we also applied and evaluated two different sets of population-based kinetic classes for SVCA4. The first set was developed by PET researchers at VUMC , and the second set was developed by the Turku PET group (TPC) using the TPC (R)-[11C]PK11195 database which partly consists of the subjects included in this article. All test-retest metrics is presented in Table 3. The VUMC classes yielded higher average BPND values, but there were no substantial differences in reliability or precision regardless of what population-based classes were used.
Convergent validity of all outcome measures
Figure 2 shows the relationships between all (R)-[11C]PK1195 outcomes derived using AIF (VT, VS, and BPND) and all outcomes derived without using AIF (BPND:SRTM-SVCA, BPND:SRTMv-SVCA, BPND:SRTM-CER, and SUV). The correlation between BPND from 2TCM and BPND from SRTM-SVCA4, SRTMv-SVCA4, or SRTM-CER was negligible to non-existent, with an explained variance (R2) < 2% for all associations. VT and VS were highly correlated (R2 = 69%), but neither showed a strong association with BPND from AIF (R2 < 9%). SUVs were not correlated to any other outcome measures (R2 < 9%).
We also fitted the 2TCM allowing vB to differ for each ROI. Ensuing BPND (k3/k4) values showed better reliability (mean ICC = 0.6) and repeatability (mean AbsVar = 23%) compared to BPND when fitted with a fixed vB, but the test-retest metrics for VT and VS were notably worse (VT mean ICC = 0.33, mean AbsVar = 21%; VS mean ICC = 0.29, mean AbsVar = 25%, see Additional file 1: Table S1). The correlation between 2TCM VT, VS, and BPND and all other BPND outcomes (SRTM-CER, SRTM-SVCA, and SRTMv-SVCA) was still low to negligible (all R2 < 9%, see Additional file 1: Figure S1).
We also evaluated different time intervals for the SUV outcome: 10–30, 20–40, and 30–50 min. No interval yielded superior test-retest metrics (ICC values ranged from 0.80 to 0.82, AbsVar 21 to 22%) compared to 40–60 min, and correlations to remaining outcomes were still negligible to low (all R2 < 12%). All SUV intervals were however strongly correlated to each other (all R2 > 95%).
The objective of this study was to examine the test-retest reliability and convergent validity of (R)-[11C]PK11195 outcomes commonly applied in clinical in vivo studies of TSPO binding. Specifically, we evaluated outcome measures of radioligand brain exposure and binding which do not make use of an arterial input function (AIF) and compared them with binding outcomes derived using an AIF (i.e., VT, VS, and BPND from the 2TCM).
There was a striking difference in magnitude between BPND values from 2TCM using an AIF and BPND values from SRTM-SVCA4 and SRTM-CER, with much higher BPND values obtained using the 2TCM compared to the other two measures. This signifies that the use of SVCA, as well as the cerebellum, for derivation of a reference TAC yields only relative or pseudo-BPND values. TSPO is expressed throughout the brain, and specific binding is to be expected in every voxel [18, 30]. Hence, it is unlikely that SVCA4 or the cerebellum can be used to establish a TAC that reflects a true reference, devoid of TSPO, for (R)-[11C]PK11195.
In general, all (R)-[11C]PK11195 outcome measures analyzed in this study showed poor to moderate reliability. For the whole GM ROI, only SUV and VT showed acceptable reliability (ICC > 0.65) . Assuming that the true TSPO concentration is stable between PET examinations, an ICC of 0.5 suggests that as much of the variance in the sample is attributable to true signal as can be attributed to measurement error and noise. All outcomes derived without the use of an AIF showed ICC values around or below 0.5, suggesting poor reliability for these measures. SRTM with the cerebellum as reference region showed the largest imprecision and MD. These results suggest that a change in BPND from SRTMCER would need to be, on average, larger than 10 times the mean in order to detect a true difference between two measurements of the same subject. In comparison, a change in VS of (on average) 40% would be necessary to detect a difference that is not only due to noise. The lack of reliability and precision for BPND from the cerebellum and SVCA can likely be explained by having a similar (low) specific to non-specific binding ratio in both the target and pseudo-reference regions, leading to TACs which are similar in both shape and magnitude. This yields BPND values which are close to zero (or negative) and which are sensitive to even small amounts of measurement error.
Another limitation when using the cerebellum as reference is that it requires researchers to establish significant equivalence [32, 33] in reference region-specific binding between the groups which are being compared. A non-significant difference between groups does not translate into evidence in favor of an absence of a difference , contrary to conclusions sometimes drawn in literature.
VT, VS, and BPND derived from 2TCM showed little to no correlation with BPND derived using outcomes without an AIF. This indicates that BPND from the reference input models have little to no convergent validity in relation to binding outcomes from AIF, and vice versa. Hence, if either VT, VS, or BPND derived using an AIF is to be considered to be at least moderately associated with specific TSPO binding, then BPND derived without the use of AIF cannot be considered valid. However, BPND from AIF also produced low ICC values and a negligible association with VT and VS, suggesting that this outcome is also unreliable and unstable. SUVs showed the highest average reliability but were not correlated with any other outcome measures.
In healthy control subjects, a large portion of the (R)-[11C]PK11195 signal consists of non-specific binding and unbound radioligand, as determined by blocking studies showing BPND values in the range of 0.8–0.9 . As described, a low signal for specific binding in healthy controls may partly explain the low reliability observed in this study. In comparison, much higher reliability has been shown for SVCA in patients with Alzheimer’s disease  where glial cell markers are known to be elevated based on postmortem studies . This gains partial support from the fact that second generation TSPO tracers, which show higher specific binding , also display higher ICC values in healthy control subjects . It is hence likely that (R)-[11C]PK11195 outcomes would show higher reliability in clinical populations with a significant increase in brain TSPO. However, this would also imply that differences between patients and controls would need to be very large in order to be detectable. While such effects may be present in some patient groups, such as patients with stroke , caution is advised for disorders for which increases in TSPO might be more subtle.
One caveat with this study is that the kinetic classes used for the SVCA method are sensitive to differences in the PET system, such as system type and acquisition protocol. It cannot be excluded that other scanners might have shown higher test-retest values for the SVCA outcomes, using the same design. Importantly, the 6-week interval between PET measurements in this study means that TSPO levels may change from test to retest. This, in turn, would lead to lower reliability and precision. However, since many clinical studies aim to evaluate longitudinal interventions or correlate (R)-[11C]PK11195 outcomes with more stable independent variables, this interval mimics that of realistic and relevant designs of PET studies. In addition, the time between measurements also should not impact the relative reliability between different outcome measures of specific binding (such as VS and BPND), nor does it affect the evaluation of convergent validity.
The results from this study suggest that caution is warranted for the application and interpretation of (R)-[11C]PK11195 BPND obtained using 2TCM or BPND from kinetic models using the cerebellum or SVCA4 as reference. VT and VS should likely be preferred over BPND from 2TCM, since they exhibited higher reliability and precision. However, the negligible correlations of VT and VS to SUVs are concerning and not fully understood. One explanation might be that brain SUV values are sensitive to changes in peripheral binding of TSPO , while AIF-based outcomes are not. This hypothesis warrants further investigation in future studies.
Two-tissue compartment model
Absolute variability (or test-retest variability)
Arterial input function
- BPND :
Non-displaceable binding potential
Intra-class correlation coefficient
Minimum detectable difference
Magnetic resonance imaging
Positron emission tomography
Region of interest
Standard error of measurement
Simplified reference tissue model
Simplified reference tissue model with additional modeling of vB
Standardized uptake values
Supervised cluster analysis
Translocator protein 18 kDa
Radioactivity contribution from whole blood
- V S :
Specific distribution volume
- V T :
Total distribution volume
Venneti S, Lopresti BJ, Wiley CA. Molecular imaging of microglia/macrophages in the brain. Glia. 2013;61:10–23 Wiley Online Library.
Politis M, Su P, Piccini P. Imaging of microglia in patients with neurodegenerative disorders. Front Pharmacol. 2012;3:96 Frontiers Media SA.
Innis RB, Cunningham VJ, Delforge J, Fujita M, Gjedde A, Gunn RN, et al. Consensus nomenclature for in vivo imaging of reversibly binding radioligands. J Cereb Blood Flow Metab. 2007;27:1533–9.
Van Der Laken CJ, Elzinga EH, Kropholler MA, Molthoff CFM, van der Heijden JW, Maruyama K, et al. Noninvasive imaging of macrophages in rheumatoid synovitis using 11C-(R)-PK11195 and positron emission tomography. Arthritis Rheumatol. 2008;58:3350–5 Wiley Online Library.
Kropholler MA, Boellaard R, Elzinga EH, Van der Laken CJ, Maruyama K, Kloet RW, et al. Quantification of (R)-[11 C] PK11195 binding in rheumatoid arthritis. Eur J Nucl Med Mol Imaging. 2009;36:624–31 Springer.
Schuitemaker A, Van Berckel BNM, Kropholler MA, Kloet RW, Jonker C, Scheltens P, et al. Evaluation of methods for generating parametric (R)-[11C] PK11195 binding images. J Cereb Blood Flow Metab. 2007;27:1603–15 SAGE publications Sage UK: London, England.
Imaizumi M, Briard E, Zoghbi SS, Gourley JP, Hong J, Musachio JL, et al. Kinetic evaluation in nonhuman primates of two new PET ligands for peripheral benzodiazepine receptors in brain. Synapse. 2007;61:595–605 Wiley Online Library.
Turkheimer FE, Edison P, Pavese N, Roncaroli F, Anderson AN, Hammers A, et al. Reference and target region modeling of [11C]-(R)-PK11195 brain studies. J Nucl Med Soc Nuclear Med. 2007;48:158–67.
Yaqub M, Van Berckel BN, Schuitemaker A, Hinz R, Turkheimer FE, Tomasi G, et al. Optimization of supervised cluster analysis for extracting reference tissue input curves in (R)-[11C] PK11195 brain PET studies. J Cereb Blood Flow Metab. 2012;32:1600–8 SAGE Publications Sage UK: London, England.
Lammertsma AA, Hume SP. Simplified reference tissue model for PET receptor studies. NeuroImage. 1996;4:153–8.
Van Berckel BN, Bossong MG, Boellaard R, Kloet R, Schuitemaker A, Caspers E, et al. Microglia activation in recent-onset schizophrenia: a quantitative (R)-[11 C] PK11195 positron emission tomography study. Biol Psychiatry. 2008;64:820–2 Elsevier.
Parbo P, Ismail R, Hansen KV, Amidi A, Mårup FH, Gottrup H, et al. Brain inflammation accompanies amyloid in the majority of mild cognitive impairment cases due to Alzheimer’s disease. Brain. 2017. https://doi.org/10.1093/brain/awx120.
Rissanen E, Tuisku J, Rokka J, Paavilainen T, Parkkola R, Rinne JO, et al. In vivo detection of diffuse inflammation in secondary progressive multiple sclerosis using PET imaging and the radioligand 11C-PK11195. J Nucl Med. Soc Nuclear Med. 2014;55:939–44.
Folkersma H, Boellaard R, Yaqub M, Kloet RW, Windhorst AD, Lammertsma AA, et al. Widespread and prolonged increase in (R)-11C-PK11195 binding after traumatic brain injury. J Nucl Med. Soc Nuclear Med. 2011;52:1235–9.
Van Der Doef TF, De Witte LD, Sutterland AL, Jobse E, Yaqub M, Boellaard R, et al. In vivo (R)-[11C]PK11195 PET imaging of 18kDa translocator protein in recent onset psychosis. NPJ Schizophr. 2016;2:16031 Nature Publishing Group.
Schuitemaker A, van der Doef TF, Boellaard R, van der Flier WM, Yaqub M, Windhorst AD, et al. Microglial activation in healthy aging. Neurobiol Aging. 2012;33:1067–72 Elsevier.
Kumar A, Muzik O, Shandal V, Chugani D, Chakraborty P, Chugani HT. Evaluation of age-related changes in translocator protein (TSPO) in human brain using 11 C-[R]-PK11195 PET. J Neuroinflammation. 2012;9:232 BioMed Central.
Doble A, Malgouris C, Daniel M, Daniel N, Imbault F, Basbaum A, et al. Labelling of peripheral-type benzodiazepine binding sites in human brain with [3H] PK 11195: anatomical and subcellular distribution. Brain Res Bull. 1987;18:49–61 Elsevier.
Kropholler MA, Boellaard R, Schuitemaker A, Folkersma H, van Berckel BNM, Lammertsma AA. Evaluation of reference tissue models for the analysis of [11C](R)-PK11195 studies. J Cereb Blood Flow Metab. 2006;26:1431–41 SAGE publications Sage UK: London, England.
Holmes SE, Hinz R, Drake RJ, Gregory CJ, Conen S, Matthews JC, et al. In vivo imaging of brain microglial activity in antipsychotic-free and medicated schizophrenia: a [11C](R)-PK11195 positron emission tomography study. Mol Psychiatry. 2016;21:1672–9 Nature Publishing Group.
Di Biase MA, Zalesky A, O’keefe G, Laskaris L, Baune BT, Weickert CS, et al. PET imaging of putative microglial activation in individuals at ultra-high risk for psychosis, recently diagnosed and chronically ill with schizophrenia. Transl Psychiatry. 2017;7:e1225 Nature Publishing Group.
Holmes SE, Hinz R, Conen S, Gregory CJ, Matthews JC, Anton-Rodriguez JM, et al. Elevated translocator protein in anterior cingulate in major depression and a role for inflammation in suicidal thinking: a positron emission tomography study. Biol Psychiatry. 2018;83:61–9 Elsevier.
Su Z, Herholz K, Gerhard A, Roncaroli F, Du Plessis D, Jackson A, et al. [11C]-(R) PK11195 tracer kinetics in the brain of glioma patients and a comparison of two referencing approaches. Eur J Nucl Med Mol Imaging. 2013;40:1406–19 Springer.
Kropholler MA, Boellaard R, Schuitemaker A, van Berckel BNM, Luurtsema G, Windhorst AD, et al. Development of a tracer kinetic plasma input model for (R)-[11C] PK11195 brain studies. J Cereb Blood Flow Metab. 2005;25:842–51 SAGE publications Sage UK: London, England.
Jučaite A, Cselényi Z, Arvidsson A, Ahlberg G, Julin P, Varnäs K, et al. Kinetic analysis and test-retest variability of the radioligand [11C](R)-PK11195 binding to TSPO in the human brain - a PET study in control subjects. EJNMMI Res. 2012;2:15.
Boellaard R, Turkheimer FE, Hinz R, Schuitemaker A, Scheltens P, van Berckel BNM, et al. Performance of a modified supervised cluster algorithm for extracting reference region input functions from (R)-[11 C] PK11195 brain PET studies. Nucl Sci Symp Conf Rec 2008 NSS’08 IEEE. 2008:5400–2 IEEE. https://doi.org/10.1109/NSSMIC.2008.4774453.
Tomasi G, Edison P, Bertoldo A, Roncaroli F, Singh P, Gerhard A, 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;49:1249–56.
Weir JP. Quantifying test-retest reliability using the intraclass correlation coefficient and the SEM. J Strength Cond Res. 2005;19:231–40 LWW.
Padfield D, Matheson G. nls.multstart: robust non-linear regression using AIC scores. 2018. Available from: https://cran.r-project.org/package=nls.multstart
Farde L. TSPO binding may also represent ‘resting’microglia. Clin Transl Imaging. 2015;3:491–2 Springer.
Portney LG, Watkins MP. Foundations of clinical research: application to practice. 3rd ed. Stamford: Lange; 2009.
Schuirmann DJ. A comparison of the two one-sided tests procedure and the power approach for assessing the equivalence of average bioavailability. J Pharmacokinet Pharmacodyn. 1987;15:657–80 Springer.
Lakens D. Equivalence tests: a practical primer for t tests, correlations, and meta-analyses. Soc Psychol Personal Sci. 2017. https://doi.org/10.1177/1948550617697177 SAGE Publications Sage CA: Los Angeles.
Dienes Z. Using Bayes to get the most out of non-significant results. Front Psychol. 2014;5:781 Frontiers.
Kobayashi M, Jiang T, Telu S, Zoghbi SS, Gunn RN, Rabiner EA, et al. 11C-DPA-713 has much greater specific binding to translocator protein 18 kDa (TSPO) in human brain than 11C-(R)-PK11195. J Cereb Blood Flow Metab. 2018; Available from: https://doi.org/10.1177/0271678X17699223.
Heneka MT, Carson MJ, El Khoury J, Landreth GE, Brosseron F, Feinstein DL, et al. Neuroinflammation in Alzheimer’s disease. Lancet Neurol. 2015;14:388–405 Elsevier.
Fujita M, Kobayashi M, Ikawa M, Gunn RN, Rabiner EA, Owen DR, et al. Comparison of four 11C-labeled PET ligands to quantify translocator protein 18 kDa (TSPO) in human brain: (R)-PK11195, PBR28, DPA-713, and ER176—based on recent publications that measured specific-to-non-displaceable ratios. EJNMMI Res. 2017;7:84 Available from: https://doi.org/10.1186/s13550-017-0334-8.
Collste K, Forsberg A, Varrone A, Amini N, Aeinehband S, Yakushev I, et al. Test–retest reproducibility of [11C] PBR28 binding to TSPO in healthy control subjects. Eur J Nucl Med Mol Imaging. 2016;43:173–83 Springer.
Kreisl WC, Mbeo G, Fujita M, Zoghbi SS, Pike VW, Innis RB, et al. Stroke incidentally identified using improved positron emission tomography for microglial activation. Arch Neurol. 2009;66:1288–9 American Medical Association.
We thank Jouni Tuisku from the Turku PET Center for his help with methodological considerations during the course of the study. We would like to thank the staff at the Karolinska Institutet PET Centre for their assistance.
S.C. was supported by the Swedish Research Council (Grant No. 523-2014-3467). The remaining authors have no funding information to report.
Ethics approval and consent to participate
The study was approved by the Stockholm Regional Ethical Board and carried out in accordance with the Helsinki declaration. Informed consent was obtained from all individual participants included in the study.
Consent for publication
The data collection in the original study  was funded by AstraZeneca. LF, ZC, and AJ are employees at AstraZeneca. AstraZeneca had no role in the idea behind or design of the re-analyses of data carried out in this article. The remaining authors declare that they have no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Figure S1. Relationships between all (R)-[11C]PK1195 outcome measures, where whole-blood contribution to ROI radioactivity (vB) has been fitted for each ROI. Values from both PET examinations and all regions have been pooled in each panel. Pearson’s correlation coefficients (r) and explained variance (R2) are presented in the upper diagonal. Table S1. Test-retest metrics for BPND, VS, and VT values derived using 2TCM while fitting vB for each ROI separately. (PDF 157 kb)
About this article
Cite this article
Plavén-Sigray, P., Matheson, G.J., Cselényi, Z. et al. Test-retest reliability and convergent validity of (R)-[11C]PK11195 outcome measures without arterial input function. EJNMMI Res 8, 102 (2018). https://doi.org/10.1186/s13550-018-0455-8
- Arterial input function
- Reference region
- Supervised cluster analysis