Introduction
Activated microglia play an important role in the pathophysiology of Alzheimer’s disease (AD) [
1]. Positron emission tomography (PET), using radioligands that bind to the mitochondrial 18 kDa translocator protein (TSPO), allows in vivo quantification of microglia in their activated state [
2]. In comparisons with unaffected controls, TSPO binding has been shown to be significantly higher in AD and there has been interest in using second-generation TSPO ligands, including [
11C]-PBR28, in longitudinal studies and clinical trials of immunomodulatory drugs [
3].
Although [
11C]-PBR28 is known to have a better signal-to-noise ratio and pharmacokinetic properties, compared to the first-generation TSPO tracer [
11C]-R-PK11195, its use in longitudinal studies of AD is limited by several factors [
4]. The absence of both a true reference region and an established pseudo-reference approach, such as the supervised clustering algorithm developed for [
11C]-R-PK11195, means that most studies using [
11C]-PBR28 require arterial cannulation [
5]. This procedure adds cost, is invasive and can be uncomfortable. This limits the use of [
11C]-PBR28 in studies with repeated measures and in AD patients, for whom tolerability and informed consent are significant issues. As an alternative to absolute quantification, three studies have recently demonstrated that a semi-quantitative, non-invasive approach using the standardised uptake value (SUV) or SUV ratio (SUVR) may be sensitive to changes in TSPO levels associated with neurodegenerative diseases including AD [
6‐
8]. Although the SUV/SUVR do not provide details of compartmental binding, they are easy to derive, do not require arterial sampling and can reduce time spent in the scanner—appealing properties for use in patients with dementia. For these approaches to have utility for longitudinal studies, the test-retest reliability (TRR) needs to be established over a time period meaningful in clinical trials. In this study, a TRR analysis of the three published SUV-based methods for analysing [
11C]-PBR28 data was conducted in a cohort of AD patients over a 12-week period.
Results
Sample characteristics
Five patients with high-affinity binding (HAB) genotype were included in the study (mean age 82.9 ± 4 years, four were male). Mean MMSE was 25.6 ± 1.3, and all patients were currently prescribed acetylcholinesterase inhibitors (AChEi). There were no significant differences in age or MMSE score between those included in the study and those excluded by genotype (mean age = 80.3 years ± 7.7 years, MMSE score = 25.8 ± 1.8).
Test-retest analysis
The test-retest analysis of the three methods of SUV analysis is presented below—unadjusted SUV, SUVRWB and SUVRC.
Unadjusted SUV
The mean test-retest values, absolute variability and ICC for unadjusted SUV values are presented in Table
1. There was significant mean absolute variability (mean −1.24 %, SD 7.28 %) across ROIs. However, unadjusted SUV measurements were highly reliable as indicated by the high ICC values (mean ICC 0.94). From these data, a sample size of 26 would be required to detect a 5 % within-subject change in MRD (averaged across all ROIs), and a sample size of 8 would be required to detect a 10 % change.
SUV relative to whole brain mean activity
The mean test-retest values, absolute variability and ICC for SUVR
WB are presented in Table
2. There was low absolute variability (mean −0.13 %, SD 2.47 %) across ROIs, and the SUVR
WB was highly reliable (mean ICC 0.83). Consequently, a sample size of 7 would be required to detect a 5 % within-subject change in MRD (averaged across all ROIs), and a sample size of 4 would be required to detect a 10 % change.
Table 2
Test-retest analysis of SUV relative to whole brain mean activity (SUVRWB)
Frontal lobe | 1.08 | 1.08 | 0.096 (1.42) | −0.09 (1.41) | 0.96 | 4 | 3 |
Parietal lobe | 1.03 | 1.04 | 0.96 (0.81) | −0.95 (0.80) | 0.98 | 3 | 3 |
Temporal lobe | 1.04 | 1.03 | −0.36 (0.73) | 0.37 (0.73) | 0.90 | 3 | 3 |
Occipital lobe | 1.07 | 1.06 | −0.51 (1.35) | 0.52 (1.37) | 0.91 | 4 | 3 |
Hippocampus | 0.94 | 0.95 | 0.71 (2.09) | −0.69 (2.06) | 0.97 | 5 | 3 |
Parahippocampus | 1.01 | 1.03 | 1.64 (3.07) | −1.59 (2.98) | 0.48 | 7 | 4 |
Posterior cingulate | 1.15 | 1.17 | 1.92 (1.36) | −1.89 (1.33) | 0.98 | 4 | 3 |
Amygdala | 1.07 | 1.07 | 0.35 (5.79) | −0.21 (5.77) | 0.69 | 17 | 6 |
Cerebellum | 1.17 | 1.13 | −3.24 (5.40) | 3.42 (5.74) | 0.63 | 15 | 6 |
Mean (SD) | 1.06 (0.07) | 1.06 (0.06) | 0.17 (2.45) | −0.13 (2.47) | 0.83 | 7 | 4 |
SUV normalised to cerebellar activity
The mean test-retest values, absolute variability and ICC for SUVR
C are presented in Table
3. There was considerable absolute variability (mean −3.98 %, SD 7.07 %) and SUVR
C showed poor reliability (mean ICC 0.65). Overall, a sample size of 25 would be required to detect a 5 % within-subject change in MRD (averaged across all ROIs), and a sample size of 8 would be required to detect a 10 % change. Of the three measures, the SUVR
C performed most poorly, due to high levels of variability in binding in the cerebellar grey matter reference region.
Table 3
Test-retest analysis of SUV relative to cerebellar activity (SUVRC)
Frontal lobe | 0.93 | 0.96 | 3.78 (7.63) | −3.50 (7.06) | 0.69 | 27 | 9 |
Parietal lobe | 0.89 | 0.93 | 4.62 (6.37) | −4.37 (5.94) | 0.70 | 20 | 7 |
Temporal lobe | 0.89 | 0.92 | 3.23 (5.87) | −3.05 (5.22) | 0.68 | 17 | 6 |
Occipital lobe | 0.92 | 0.94 | 3.04 (4.98) | −2.90 (4.75) | 0.71 | 13 | 5 |
Hippocampus | 0.81 | 0.84 | 4.44 (8.18) | −4.11 (7.54) | 0.77 | 31 | 10 |
Parahippocampus | 0.87 | 0.92 | 5.46 (9.51) | −5.00 (8.59) | 0.38 | 40 | 12 |
Posterior cingulate | 0.99 | 1.05 | 5.65 (7.25) | −5.31 (6.65) | 0.79 | 25 | 8 |
Amygdala | 0.92 | 0.95 | 4.18 (11.53) | −3.62 (10.51) | 0.47 | 25 | 8 |
Mean (SD) | 0.90 (0.05) | 0.94 (0.06) | 4.30 (7.66) | −3.98 (7.07) | 0.65 | 25 | 8 |
Discussion
Of the three SUV measures present in this test-retest analysis, the SUV normalised to whole brain mean activity performed most strongly. The unadjusted SUV showed higher variability, and the SUVRC showed both higher temporal variability and lower reliability. Lower reliability in SUVR methods, as measured by ICC, was most apparent in smaller regions with lower signal to noise ratios, such as the parahippocampal gyrus, which is affected early in the AD process.
Compared to published data on the TRR of [
11C]-PBR28, the SUV methods presented here perform well [
13,
14]. Park et al. reported that, over a 1.4-week period, in healthy controls and patients with multiple sclerosis, the SDs of test-retest variability of
V
T and
V
T/
f
p (derived using a multi-linear analysis) varied from 9 to 11 % and 7 to 14 %, respectively [
13]. Collste et al. [
14] reported that the mean absolute variability in
V
T in the grey matter was 18.3 ± 12.7 %, with ICC values from 0.90 to 0.94, and that, using a parametric modelling approach, variability was 17.8 ± 12.7 %. Participants were imaged either on the same day or 1 day apart. By comparison, all three SUV measures in our study performed well, over a much longer time period. These results also compare favourably with TRR analysis of the first-generation TSPO ligand [
11C]-R-PK11195 [
5,
15].
These findings should be viewed as a preliminary indicator that SUV-based methods may be suitable for use in longitudinal studies in AD. Our findings should be confirmed in larger samples, which could also include medium-affinity binders to assess the impact of genotype on longitudinal variability. Further work is also needed to confirm the relationship between SUV measures and measures of specific binding. Although there are concerns about the non-specific binding profile of [
11C]-PBR28, work by Lyoo et al. suggests that SUVR
C correlates well with absolute quantification, but more data on this is required [
8,
16]. This validation is also required for alternative SUV-derived methods, and this could be easily achieved through retrospective analysis of existing data sets. Although SUV and SUVR
WB appear to have better test-retest reliability, other factors such as sensitivity to disease states may vary between methods and should be considered when choosing a quantification method.
In conclusion, given the caveats listed above, this TRR analysis of SUV-derived measures of [11C]-PBR28 binding in AD suggests that non-invasive semi-quantitative approaches are stable and reliable over significant periods of time.
Acknowledgements
This study was funded by the National Institute for Health Research (NIHR) Dementia Biomedical Research Unit and Biomedical Research Centre at South London and Maudsley NHS Foundation Trust and King’s College London. The views expressed are those of the authors and not necessarily those of the NHS, the NIHR or the Department of Health. The first author, AN, received a salary support from the National Institute for Health Research (NIHR) Mental Health Biomedical Research Centre (BRC) at South London and Maudsley NHS Foundation Trust and King’s College London as a BRC Preparatory Clinician Fellow. MV and FT were supported by the MRC Programme Grant ‘PET Methodology’ (Ref: G1100809/1). RH is supported by the UCLH NIHR Biomedical Research Centre.
The authors thank all the clinical imaging staff at Imanova for their assistance with this study.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (
http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.